Assessment of driving factors for yield and productivity developments in crop and cattle production as key to increasing sustainable biomass potentials , driving factors , sustainability , yield

The sustainable production potential of biomass for energy and material pur-poses largely depends on the future availability of surplus agricultural lands made available through yield improvements in crop and livestock production. However, the rates at which yields may develop, and the inﬂ uence of technological, economic and institutional factors on these growth rates are key uncertainties in assessing the potentials and impacts of biomass production. This study analyzes the pace and direction of historical yield developments (1961–2010) of ﬁ ve major crops, beef and cow milk in Australia, Brazil, China, India, USA , Zambia, and Zimbabwe, and examines the driving factors behind these developments. In addition, it explores how future yields are modeled and how modeling efforts may be improved. Average yield growth rates over the investigated period ranged in most cases between 0.7–1.6% year −1 for crops, 1.0–1.5% year −1 for milk, and 0.4–0.8% year −1 for beef (relative to 2010). The role of different drivers is region speciﬁ c. Yet, supporting agricultural policies have played an important role in increasing yields in all countries, especially for crops. In cattle production, a key factor was the importance of commercial beef and milk production for the national or export market. Based on regional differences in drivers and yield developments, models that assess biomass potentials and impacts should take into account regional drivers, yield gaps, and potential policy pathways.


Introduction
The use of biomass for energy, chemicals, and materials is considered an important alternative to fossil resources (Chum et al. 2011 ;Harvey and Pilgrim 2011 ). For biomass to deliver a sizeable contribution, the availability of suffi cient sustainable and affordable biomass feedstock is crucial. Assessment studies that have evaluated the current and future global availability of biomass resources show that the largest future potential contribution can come from energy crops grown on various types of land (Hoogwijk et al. 2005 ;Smeets et al. 2007 ). The most important land type is surplus agricultural land, which can be released through increased production effi ciency of food, animal feed or pasture. There is, however, disagreement about the availability of surplus agricultural land. Key uncertainties in predicting this area of land are technological progress in agricultural production systems and the related increases in crop and livestock yields (Dornburg et al. 2010 ;Slade et al. 2011 ;Batidzirai et al. 2012 ).
Several studies have investigated the effects of yields on the availability of surplus agricultural land and biomass potentials and impacts. For example, van Vuuren et al. ( 2009 ) assessed the impact of food crop yield changes on the global woody biomass potential in 2050. They found that an additional yield improvement of 12.5% compared to the baseline scenario resulted in an increase of the biomass potential from 150 to 230 EJ. Erb et al.

ORIGINAL RESEARCH
Assessment of driving factors for yield and productivity developments in crop and cattle production as key to increasing sustainable biomass potentials ( 2009 ) found that the biomass potential in 2050 would be 79 EJ year −1 in the case of intermediate agricultural intensifi cation and humane livestock rearing and 105 EJ year −1 in the case of greater intensifi cation of crop and livestock production. 1 Dornburg et al. ( 2010 ) estimated that improvements in agricultural management could account for 140 EJ year −1 of the total biomass supply potential of 500 EJ year −1 in 2050. Slade et al. ( 2011 ) derived from a review study that more than 1 Gha of high yielding agricultural land, equal to about 20% of the global agricultural land area in 2010, could be made available for bioenergy crops in 2050 if food crop yields increase at a higher rate than food demand and if the consumption of livestock products is limited.
The degree of yield improvements also affects the environmental performance of biomass production. Without suffi cient improvements in yields, there is a large risk of direct or indirect land use change (DLUC and ILUC, respectively), which can result in high greenhouse gas (GHG) emissions (Searchinger et al. 2008 ;Laborde 2011 ). In addition, advances in agricultural production systems may also improve the performance of the agricultural sector as a whole. For example, Tilman et al. ( 2011 ) show that there is a signifi cant potential in agriculture to reduce global land clearing, GHG emissions and nitrogen use through improved technology and adaption and transfer of high-yielding technologies to underyielding regions. Also, Havlík et al. ( 2014 ) show that the transition of livestock production toward more effi cient systems would signifi cantly decrease livestock-induced GHG emissions. These emission savings are mainly a result of a reduction in land use change (Havlík et al. 2014 ).
Models that assess land availability, land use change induced by biomass demand and other impacts of biomass production, such as those used in the studies mentioned above, generally base their crop yield projections on historical developments. Many of these studies also account for (a limited number of) endogenous drivers of future yields. These factors are related to, for example, climate change (Jaggard et al. 2010 ), crop or land prices (Eickhout et al. 2008 ;Rosegrant et al. 2008 ;EPA 2010 ;Khanna et al. 2011 ) or management changes like the increased use of fertilizer and other production factors (Eickhout et al. 2008 ;Beach et al. 2010 ;Mosnier et al. 2012 ). This diversity of factors refl ects that in reality yield developments depend on numerous factors of various origins (e.g., economic, technological, and ecological). The question arises as to what role these different driving factors play, how they relate to each other and if their impact varies between regions. Moreover, productivity developments in the livestock sector have received much less attention in literature and modeling efforts than agricultural crops -despite the fact that livestock production accounts for 70% of the total agricultural land and one third of the arable land area is used for feed crop production (Steinfeld et al. 2006 ). For this sector, the lack of insight into the possibilities to increase yields, the rate at which this can be established and the role of different driving factors is even larger. Recently, de Wit et al. ( 2011 ) discussed what growth rates and maximum (sustainable) yields could be achieved in European agriculture. They assessed agricultural yield developments in the past fi ve decades and compared these to policy developments, structural changes and trends in the use of production factors (inputs). De Wit et al. found that yield developments were clearly correlated to agricultural policy, but yield growth did not always coincide with more effi cient use of inputs (de Wit et al. 2011 ). De Wit et al. focused on Europe and did not investigate other regions that are of critical importance in future biomass supply such as Latin America and sub-Saharan Africa (Hoogwijk et al. 2005 ;Smeets et al. 2007 ). Given the importance of yield projections in determining biomass supply and impacts, the aim of this study is to assess for seven countries in different world regions (i.e., Australia, Brazil, China, India, United States, Zambia, and Zimbabwe): 1 . what the historical agricultural developments and their drivers are, 2 . to what extent and at what growth rate crop and livestock product yields can improve in the future, and 3 . how different settings and drivers can infl uence future yield developments .
These insights contribute to several aspects identifi ed as a key to improving the assessment of biomass potentials and impacts, such as (1) the use of bottom-up analyses to enhance the understanding of current (agricultural) systems, options for improvement, the degree to which yields can be increased, drivers, and regional differences, and (2) a more explicit discussion of assumptions (including yields) (Batidzirai et al. 2012 ;Wicke et al. 2014 ).

Selection of agricultural products and producing countries
The potential area of surplus agricultural land is expected to be largely infl uenced by effi ciency developments in the production of major agricultural products. Therefore, the agricultural developments are assessed for fi ve crops that are most dominant in terms of global production and cultivated area: wheat, corn, rice, sugarcane, and soybean ( FAOSTAT ). In addition, for livestock, we only take into account beef and cow milk production since cattle uses most of the agricultural area for grazing. Pig and chicken production are often landless, but land is required for producing feed crops (Seré and Steinfeld 1996 ). The area of cropland needed largely depends on the crop yields, which are already taken into account in this study. Therefore, pig and chicken production are not included in the assessment. Agricultural developments are assessed in seven countries: Australia, Brazil, China, India, the United States, Zambia, and Zimbabwe (Fig. 1 ). There are two reasons for this selection. First, Brazil, China, India, and the United States are major producers of the selected crops and cattle products ( FAOSTAT ). Second, Australia, USA, Zambia, and Zimbabwe can potentially release a large area of agricultural land for biomass feedstock production (Hoogwijk et al. 2005 ;Smeets et al. 2007 ). de Wit et al. ( 2011 ) assessed agricultural developments in France, The Netherlands, Poland, and Ukraine. For comparison, we have included data about France in the results section. In addition, we compare the general fi ndings from de Wit et al. with our own results.

Historical developments in driving factors
The analysis starts with a description of the current status of agriculture in the selected countries and developments in driving factors that have taken place since 1961 (this part of the study is presented in Appendices 1-8 ). Based on literature, the drivers of yield developments are classifi ed into three types (Anderson 2010 ;Hengsdijk and Langeveld 2010 ;Neumann et al. 2010 ;Piesse and Thirtle 2010 ;Smith et al. 2010 ;de Wit et al. 2011 ): technological/ management, economic, and institutional. Economic drivers are, for example, market developments and agricultural R&D investments. Institutional drivers include agricultural policies and governance systems. The discussion of economic and institutional drivers is based on literature review. For technological/management drivers, the following are assessed: labor intensity and level of mechanization, irrigation, nutrient, and pesticide use. These indicators are derived from time series data  collected from the UN Food and Agricultural Organization statistical division ( FAOSTAT ). These statistics are aggregated on a country level, for example, annual national consumption of fertilizers (tonne year −1 ). To enable comparison of the management levels between countries, the factors are expressed in average intensity per hectare of agricultural land, for example, the national number of tractors used divided by the total area of agricultural land. For cattle, another indicator for the management level or production intensity is the proportion of ruminants to the area of meadows and pastures (hereafter the ruminant density). This is also derived from FAO statistics. It is not chosen to evaluate the cattle density, because this neglects the importance of other ruminants in the occupation of meadows and pastures for grazing Figure 1 . Selected countries and agricultural products. *France was earlier assessed by (de Wit et al. 2011 ), but for comparison, data for France are also presented in this study. and hay. As a result of the feed requirements varying between ruminant species, the number of ruminants is expressed in livestock units (LU), where one unit represents the energy requirements for maintenance and production of a typical cow in North America. The livestock unit coeffi cients are obtained from the FAO ( 2011 ). The ruminant density is then calculated as the number of livestock units per hectare of meadows and pastures. The ruminants included are: buffaloes, camel, cattle, goats and sheep. The area of meadows and pastures consists of the total land area available for both for grazing and for the production of conserved forages. This approach thus also accounts for systems that combine grazing and confi nement. In this study, we only consider a limited number of drivers that can be actively steered. Literature shows that more factors can infl uence yields, see the discussion.

Historical yield and productivity developments
To assess historical yield trends and yield growth rates for the selected products and countries, time series data  are collected from FAOSTAT . The crop yield is defi ned as the annual production quantity per hectare of area harvested (tonne ha −1 year −1 ); the beef yield is given in terms of carcass weight (kg animal −1 ); the milk yield is the annual milk production per cow (kg animal −1 year −1 ). All numbers are national averages. The average beef and milk production per animal, however, are not the best indicators to study developments in livestock product yields. Beef and milk production can take place in different production systems ranging from pastoral to landless. But intensifi cation does not always lead to higher beef or milk production per animal (e.g., because faster weight gain leads to shorter lifespans). Therefore, a better parameter for milk and beef yields would be the feed conversion effi ciency (FCE, kg animal product per kg feed intake). The use of the FCE, however, has also limitations. These are considered in the discussion.
Average annual yield growth rates are obtained by applying linear regression to the historical yield data and are presented per product, per country, per decade and for the entire period. Growth rates are both expressed in absolute growth per year (e.g., t ha −1 year −2 ) and in percentage per year (simple annual growth rate, relative to the initial year). Temporal shifts are identifi ed for each product on country level and differences between products within a country are described. Explanations are sought by comparing the observed changes with the technological/ management, economic, and institutional developments. In addition, developments in productivity of the total agricultural sector and of the total livestock sector are assessed and discussed. This productivity is defi ned as the proportion of aggregated outputs to aggregated inputs (output-input ratio). To derive the aggregated inputs and outputs, the trend of all inputs and outputs in physical units is calculated as an index (base year 1961). From these indices, an (unweighted) average of all inputs and all outputs is calculated for each year. For the agricultural sector. the included inputs are: agricultural land, fertilizer and tractors; the outputs are: crops, meat, milk, and eggs. For the livestock sector, the inputs are: feed crops, meadow, and pasture land; the outputs are: meat, milk, and eggs.

Future yield projections and the role of driving factors
Yield growth rates from projections in literature and models are compared to linear extrapolation of historical trends. It is discussed how yield projections are defi ned and how they can be improved based on the fi ndings on historical driving factors. For each country, key factors are identifi ed that may stimulate or limit future yield developments.
To better understand what possible pathways could be defi ned for future yield developments, also the magnitude of yield gaps is taken into account. The yield gaps are derived from current yield levels and data on maximum attainable rain-fed or irrigated yields in 2020 as derived from the Global Agro-Ecological Zones database for the IPCC SRES B1 Scenario from the Australian Commonwealth Scientifi c and Research Organization (CSIRO) Mark 2 Mode ( GAEZ Global Agriecological Zones ).

Yield and productivity developments
For each country, historical developments in agricultural inputs and yields are assessed. For each country, the status of agriculture and the developments in the different driving factors are discussed in detail in Appendices 1-8 . Here a synthesis of historical yield developments and driving factors is presented for the case of Zimbabwe. Syntheses for the other countries can be found in Appendices 2-7 .
The key fi ndings for each country are presented and compared after the example of Zimbabwe.

Example: developments and driving factors in Zimbabwe
Zimbabwe ' s fi rst green revolution by commercial farmers (Eicher 1995 ;Langyintuo and Setimela 2009 ) is clearly represented by the increase of irrigated area and fertilizer use in the 1960s and early 1970s (Fig. 2 ). The yields of corn and soybeans only started to improve from the presenting a selection of the data to reveal differences between the countries at the lowest levels of tractor use. Note: when no data are shown for a certain country and/or year, no data is available for this country or year. Note: for tractors, the capacity or size of machinery is not taken into account.  1960 1970 1980 1990 2000 2010 1960 1970 1980 1990 2000 2010 1960 1970 1980 1990 2000 2010 1960 1970 1980 1990 2000 2010 1960 1970 1980 1990 2000 2010 1960 1970 1980 1990 2000 2010 1960 1970 1980 1990 2000 2010 1960 1970 1980 1990 2000 2010 Irrigated area (%) 0 0 second half of the 1960s (Fig. 3 ), which coincides with the shift from tobacco production to other crops because of export sanctions imposed in 1965 (Whitlow 1988 ;Eicher 1995 ) (also see Appendix 8 ). During the 1970s, corn yields declined again, while soy yields only fell down in 1979. In addition, irrigation levels stagnated and fertilizer use dropped in the 1970s. Thus, management conditions seem to be affected by the civil war (Whitlow 1985 ) and corn production suffered more from this than soybean cultivation. Sugarcane yields appear to be even less affected by economic and political changes in the 1960s and 1970s. Apart from signifi cant fl uctuations, sugarcane yields have increased from 1960 until 1986. The improvement rate of 0.9% year −1 in these years, however, is signifi cantly lower compared to a growth rate of 3.1% year −1 for corn and 16.9% year −1 for soybeans between 1960 and 1980 (all relative to 1961), also see Table 1 .   1960 1970 1980 1990 2000 2010 1960 1970 1980 1990 2000 2010 1960 1970 1980 1990 2000 2010 1960 1970 1980 1990 2000 2010 1960 1970 1980 1990 2000 2010 Yield (  The introduction of smallholder support in the early 1980s led to a second green revolution (Eicher 1995 ), but this is not clearly refl ected in the statistics. A major factor is the severe drought in 1983, resulting in a signifi cant drop of corn and soybean yields. Good climate conditions in 1985 led to high yields (Eicher 1995 ). But due to the reduction of smallholder support, and maybe the decline of agricultural R&D as well (Eicher 1995 ), fertilizer use and crop yields declined in the late 1980s and the 1990s. Remarkably, the area under irrigation increased in the same period. Due to a severe drought, crop yields plummeted in 1992 (Eicher 1995 ). With the fast-track land reform in 2000 (Matondi, 2012a ), yields and input levels dropped further and continued declining in the following years. Overall, crop yields and also agricultural productivity (Fig. 4 )  The average annual yield change in terms of percentage is given relative to the fi rst year of the selected period, that is, the average growth rate is expressed as a percentage of the estimated yield in the fi rst year of selected period (e.g., from 1971-1980, the average annual growth of corn yield in Brazil was 1.5% of the yield level in 1971 as derived from linear regression). 4 Note that the yield growth rates per decade are calculated for a limited time period of 10 years. This means that the choice for a certain timeframe and outliers in the data can have signifi cant infl uence on the result of the linear regression. The use of longer timeframes may show a very different trend in yield development. 5 Relative to 2010. Considering cattle production, beef yields were stable in the 1960s and early 1970s, but declined during the ci vil war in the late 1970s (see Fig. A9 in Appendix 9 ). In the 1980s and 1990s, the yields improved at an average rate of 1.8% year −1 (relative to 1981), but dropped temporarily during the drought of 1992. According to the FAO statistics, beef yields have stagnated since the hyper-infl ation and outbreak of foot-and-mouth disease (Marquette 1997 ;Matondi, 2012b ). Between 1960 and 1990, milk yields increased at a very low rate of 0.2% year −1 (relative to 1961). Thus, it seems that technological improvements were limited while economic and political changes did not signifi cantly affect milk yields either. After a small drop in 1992, yields peaked in 1993 and have stabilized since the late 1990s.

Summary and comparison between countries
Over the past fi ve decades, most crop yields showed an upward trend (Table 1 ). The yield growth rates, however, varied signifi cantly between regions. Average yield growth rates over the period investigated  ranged in most cases between 0.7-1.6% year −1 for crops, 1.0-1.5% year −1 for milk and 0.4-0.8% year −1 for beef (all relative to 2010). Highest rates were found for wheat in China (1.8% year −1 ), milk in India (1.5% year −1 ), and beef in France (0.9% year −1 ). The lowest rates for a crop are found for sugarcane (−0.3 to 1.0% year −1 ), for any one country the rates are lowest for Zimbabwe (−1.6 to 0.7% year −1 ). For comparison, in the European countries studied by de Wit et al. ( 2011 ), the average growth rates of wheat   In this study, the most observed trend over fi ve decades for crop yield growth is linear. This is in accordance with other studies (Hafner 2003 ;Ray et al. 2013 ;Fischer et al. 2014 ). Yet, in each case, the analysis revealed periods during which yields improved at a higher rate compared to the long-term average as well as periods during which yield growth rates were lower than this average. In each case, technological as well as economic and institutional factors have played a role and these drivers often infl uenced each other. Yet, the importance and the effect of a driving factor varied from case to case. Table 2 gives an overview of the most important factors behind yield and productivity developments in each country (for more details, see Appendices 2-8 ; for France see de Wit et al. ( 2011 )).
Improvements in agricultural technology and management have often led to considerable yield growth (Figs 2 and 3 ). Especially in China and India, large-scale adoption of new technologies (including high-yielding crops) resulted in high average yield growth rates (Appendices 4 and 5 ). In France and the United States, improved technologies resulted in considerable absolute yield improvements (Appendix 6 ). The technological improvements, however, included a signifi cant increase in the use of inputs like fertilizer and often caused a decline in agricultural productivity (Fig. 5 ). In the cattle sector, yield improvements were often achieved through the increased use of feed crops. In Australia and Zimbabwe, however, the consumption of feed crops grew faster than the production of meat and milk, which led to a reduction in the productivity of the livestock sector. Other countries, like India and the USA were able to compensate for the higher input levels by increasing the production output levels of the livestock sector at a similar or even higher rate.
Economic factors often play a vital role in the improvement of agricultural technology. Investments in R&D enabled the development of new technologies. In most countries, these were mainly public investments. In the USA, also private investments were very important (Appendix 6 ). These investments had already started in the period of industrial and agricultural protectionism. Only in Australia, industrial protectionism indirectly biased the agricultural sector and hindered improvements in production practices and crop yields (Appendix 2 ). The introduction of economic liberalization provided an incentive for many farmers to (further) improve yields and increase or stabilize agricultural productivity. In Australia, yields and yield improvement rates improved quickly after liberalization started. In Brazil and the USA, yield improvement rates reduced in the fi rst instance but increased again after about ten years (Appendices 3 and 6 ). Agricultural production in China diversifi ed after the reforms and yield improvements of predominant crops slowed down (Appendix 4 ). Commercial farmers in Zambia profi ted from liberalization as they were able to improve soybean yields at high rates, while corn yields of smallholders decreased (Appendix 7 ). For the cattle sector, the importance of commercial beef and milk production for the domestic or export market was found to be a key factor for yield improvements. In Australia and the USA, such markets already existed during the period of protectionism, while the beef market in Brazil has especially grown after economic liberalization (Appendices 2 , 3 and 6 ). In France (and the EU as a whole), dairy markets got saturated and a quota on milk production was introduced. The number of dairy cows reduced signifi cantly, but milk yields continued to increase through improved management ( Huyghe 2012 ). This, however, led to a reduction in livestock productivity in terms of the outputinput ratio.
In accordance with the fi ndings of de Wit et al. ( 2011 ) for European countries, policies are found to be an important instrument in steering changes in the agricultural sector in the seven countries investigated in the present study. New technologies could be adopted by farmers because of farmer support programs, for example, in India (Appendix 5 ). Market liberalization policies created new markets for agricultural products, for example, in Australia (Appendix 2 ). In some cases, yield improvements have been attained by policies that were focused on a specifi c commodity: for example, in Zambia, the focus of policies on corn production during a long period resulted in signifi cantly higher yield growth rates for corn compared to other crops (Appendix 7 ). In Brazil, the ProÁlcool program positively affected sugarcane yields (Appendix 3 ). The import substitution policy for edible oils in India, especially stimulated the increase of soybean yields. In contrast to the successful implementation of policies in the above examples, Zimbabwe also shows the impact of a lack of good governance and stimulating policies. A civil war in the 1970s and economic reforms around the 1990s disrupted agricultural production and the economy. Due to Driving Factors for Yield Developments the long-term unstable situation, crop yields and agricultural productivity have fl uctuated heavily. In addition to agricultural policies aimed at economics and production, most countries have also introduced agri-environmental policies which aimed at, for example, enhanced quality of degraded agricultural lands (Australia, China, Zambia, Zimbabwe), balanced use of inputs (China, France) and controlled use and management of natural resources (Australia, India, the United States). In the USA, China, and Zambia, this led to improved agricultural productivity (Appendices 4 , 6 , and 7 ). In Australia, the productivity did not improve considerably compared to previous years (Appendix 2 ). In India, the productivity continued to decline due to weak enforcement of agri-environmental policies (Appendix 5 ).

Crops
Models that assess biomass potentials and/or impacts of biomass production apply either only exogenous yield projections (determined by factors outside the model) or a combination of exogenously and endogenously (determined by the model based on internal factors) defi ned yield projections. The exogenous yield projections are based on historical trends. As mentioned in the previous section, the analysis of historical yield developments shows that the most observed trend for crop yield growth is linear. Yet, longer historical time series show that in, for example, the USA crop yield growth has not always followed the current linear trend (see Fig. A11 in Appendix 9 ). Also, over shorter time frames, variability in the trend is found with periods of decline, stagnation and/or strong growth. Therefore, yield projections based on historical trends depend on the historical time frame taken into account. For example, Fischer et al. ( 2014 ) fi nd global yield growth rates of 1.0% year −1 for wheat, rice, and soybean and 1.5% year −1 for corn based on the linear trend for 1991-2010. For the period 1961-2010, the present study fi nds global growth rates that are slightly higher for wheat, rice, and soybean (1.1-1.3% year −1 ) and lower for corn (1.3% year −1 , all relative to 2010), see Table A1 in Appendix 9 . Although these differences seem small, they may have considerable impact on future biomass potentials and impacts. This is illustrated by Fischer et al. ( 2014 ) who state that, in order to meet the projected food demand in 2050 with limited increase of real prices of crops, the minimum global yield growth rate for staple crops between 2010 and 2050 is 1.1% year −1 relative to 2010 (Fischer et al. 2014 ). Higher growth rates of, for example, 1.3% year −1 are preferred to account for factors that may infl uence supply and demand of crops, including increasing biofuel demand (Fischer et al. 2014 ). In model assessments, it is therefore important to make explicit what historical time frame is considered to defi ne exogenous yield projections. Although exogenous yield projections are based on historical trends, some models assume extrapolation of this trend (e.g., in (Jaggard et al. 2010 )), while most models assume the overall future yield growth to slow down compared to the historical trend (e.g., in (Laborde 2011 ;OECD 2012 )). Van Dijk and Meijerink ( 2014 ) Figure 5 . Comparison of developments in productivity (output-input ratio) of the agricultural and livestock sector in the seven selected countries. The input-output ratio is indexed to 100 for the year 1961. This means that when the ratio is higher than 100 in 1 year, the productivity has improved compared to 1961; when the ratio is lower than 100, the productivity has declined. For the agricultural sector the included inputs are: agricultural land, fertilizer and tractors; the outputs are: crops, meat, milk and eggs. For the livestock sector, the inputs are: feed crops, meadow and pasture land; the outputs are: meat, milk and eggs. give several reasons for assuming decreasing yield growth. First, the opportunities for increasing yields and exploiting existing yield gaps are more and more exhausted (Searchinger et al. 2013 ). Also, investments in agricultural R&D have declined and considerable socioeconomic constraints in many developing countries are considered to remain a limiting factor for yield growth (Alston et al. 2009 ;McIntyre et al. 2009 ;Alexandratos and Bruinsma 2012 ). Although these motivations are reasonable, these are mainly expectations about how different factors are likely to develop. There may, however, also be other possible pathways. Indeed, the present study shows that in the past 50 years, significant developments in technology, for example, mechanization, fertilizer use, and crop breeding, have driven yield improvements. But although these technologies are wide-spread now, there still exist considerable differences in technology level and yield gaps between regions (see first part of the results and Table A2 in Appendix 9 ). The present study and the study by de Wit et al. ( 2011 ) show that in regions where historical yield growth was high, the development and adoption of new technologies was primarily driven by policies (e.g., subsidies for farmers, trade liberalization and public investments in R&D). In regions where there is still room for considerable yield improvements, stimulating policies could thus play a vital role in materializing this potential. Similarly, other factors could also have a significant effect on future yield developments. The presence of different potential pathways shows that, in the assessment of future biomass potentials and the impacts of biomass production, it is important to investigate different scenarios and to include various endogenous driving factors.
To determine the endogenous yield projections, models relate yield change to driving factors like land or crop prices, climate change, or management. According to Dietrich et al. (2014), technological change is considered to be the key driver for yield change. There is, however, little consensus about the drivers of technological change and the infl uence of these drivers on yield change (Dietrich et al. 2014 ;Robinson et al. 2014 ). Therefore, the number of endogenous factors is generally limited in models, often to only one or two factors. As a result, the different technical, economic, and institutional driving factors are not covered well. Furthermore, the fi ndings from the historical analysis make clear that future yield developments largely depend on how the different driving factors develop in each region. As the number of endogenous factors is generally limited, models do not properly distinguish different driving factors between regions. Important examples of how different drivers affect yield growth potentials are the following: • In the case of corn in the USA, current yields have attained almost 80% of the maximum attainable yield 2 (see Table A2 on yield gaps, Appendix 9 ). Thus, it is likely that the current technological limits will constrain and slow down yield growth in the near future. Continuation of the historical yield improvement trend would require signifi cant technological progress , for example, increase in the potential agroclimatic yield through biotechnology. • On the other side of the spectrum, yields of rice and soy in India and corn and soy in Zambia and Zimbabwe are less than 40% of the maximum attainable yield. In various other cases, the yield gap is smaller but still leaves room for considerable yield improvements as well.
In such situations, the historical analysis shows that accelerated yield growth compared to the longer term trend is possible under favorable circumstances with regard to, for example, governance. • Zambia and Zimbabwe are examples of cases where stable agricultural and trade policies are needed to improve and support the agricultural sector and market and the economy in general. Under such conditions, farmers may be able to adopt improved technologies and management practices. As more advanced technologies and practices already exist, farmers could realize a signifi cant acceleration in yield growth compared to the average trend over the past fi ve decades. • In Southern India, rice yields could be increased signifi cantly if management conditions and market access would be improved (see Appendix 5 on agricultural characteristics). In Northern India, rice (and also wheat) is mainly produced on irrigated lands and yields are higher compared to Southern India. Ground water depletion, however, poses a risk to future yield improvements. • An important measure to attain yield improvements in an environmentally sound way is sustainable agricultural intensifi cation . In Australia, for example, management levels have been low relative to most other countries. The same is true for yield growth rates. Although intensifi cation could have signifi cant environmental impact, it is found that this can also be realized while improving the productivity, that is, increasing the crop production (output) per unit resource use (input), and reducing negative effects like emissions and water pollution (see also (Fischer et al. 2014 ;Hochman et al. 2013 ;de Wit et al. 2014 )).
Thus, for future modeling work, detailed regional assessment of the most important driving factors for yield development and the implementation of more drivers are needed. For this purpose, Table 3 identifi es some key drivers that may either limit or stimulate future yield improvements in each country assessed in this study.  ( 2014 ) show that in economic models, policies and institutions are included, but it seems that their effect on yields is not considered yet. Given our fi ndings, it is important to include policies as a driving factor for yield changes in model assessments. For example, Dietrich et al. ( 2014 ) have attempted to implement endogenous yield change related to investments in R&D. Linking yield developments to R&D and other policy-related drivers could be used to defi ne scenarios for different policy pathways and to evaluate their impact on yield changes. In addition, yield gap fi gures are a good indicator both for the degree of technological progress that can still be attained, and for the potential yield growth rates. It is useful to apply yield gaps in the models to defi ne the yield development projections. For that, more research is needed on, for example, the (crop and region specifi c) correlation between yield gaps and yield growth rates. In the historical analysis, it was also found that yields have often fl uctuated in the past and this affected the average yield growth rate. In current assessment models, projections are based on historical trends and the infl uence of fl uctuations on the long-term trend is neglected. The regional identifi cation of key drivers for yield developments could include an assessment of risk factors for yield fl uctuations (e.g., the occurrence of extreme climate conditions), which can be taken into account in the yield projections.
A comparison of yield projections from global outlook studies in van Dijk and Meijerink ( 2014 ) shows that the projected yield growth varies signifi cantly and depends on the underlying assumptions made. The infl uence of underlying assumptions on yield projections is also illustrated by a comparison of yield projections from the integrated assessment model IMAGE (which is used for the assessment of biomass potentials and impacts; see van Vuuren et al. ( 2009 )) and the economic model MIRAGE models (which is used for analyses on, land use change induced by biofuel targets; see Laborde ( 2011 )). In Table A1 (Appendix 9 ), growth rates derived from the yield projections in IMAGE and MIRAGE are compared to the extrapolation of historical trends. Both models combine exogenous yield projections with endogenously determined yield changes. The exogenous yield projections used in IMAGE and MIRAGE assume that, on the global level, yield growth rates will slow down compared to the historical linear trend (Laborde 2011 ;OECD 2012 ). Nevertheless, the two models do not always agree on whether yields in a certain region or even globally will improve at a pace higher or lower compared to the linear growth trend. Large differences in projections between the models are found for corn and soybean in Brazil and rice in China. Also, the projected global growth rates from MIRAGE for corn, rice, and soybean are higher compared to the projections based on linear extrapolation of the historical trends from 1961-2010. In IMAGE, the global projections for wheat, corn, rice, and soybean result in lower yield growth rates compared to linear extrapolation. It is most likely that these contrasting results can be explained by differences in how endogenous yield changes are modeled. The insights from this and other studies would help to make the underlying assumptions for endogenous yield projections more explicit and detailed, and help to assess how yield projections are infl uenced by different assumptions. Continuation of unstable economic situation, climate variability Re-establishment of (beef) export markets: improvement of knowledge and production of smallholders The factors are identifi ed based on the historical assessment presented in the fi rst part of the results section.
In the introduction, several studies were mentioned that assess the infl uence of increased yields on the biomass potential. Van Vuuren et al. ( 2009), Erb et al. ( 2009 and Dornburg et al. ( 2010 ) used yield projections from the FAO (Bruinsma 2003 ), the presented yield projections from the IMAGE model are in line with these projections (OECD 2012 )). Van Vuuren et al. ( 2009 ) andDornburg et al. ( 2010 ) take this FAO scenario as baseline and assess the extra biomass potential from additional yield increases. Erb et al. ( 2009 ) however, assume that the FAO scenario represents a high intensifi cation scenario and baseline yield improvements are lower. This shows that the perception of the baseline varies. This again underlines the importance to make assumptions more explicit. In addition, it is important to discuss each scenario and address under what conditions the projected yields and the resulting biomass potentials and impacts can be attained; the degree to which investments have to be increased, or required changes in policy. This is considered to be highly valuable for decision making.

Beef and milk
As opposed to historical yield growth trends for crops often being linear, for beef and milk production, we only found linear trends for Australia, India, and the United States; these are countries where we found that yields had significantly improved over a longer timeframe because of the existence of a commercial market. The absence of a yield trend in the past makes it more difficult to define yield growth scenarios for the future. Also, compared to crops, less information can be found about the projections used in the models. Several studies, e.g. IMAGE (Bouwman et al. 2005 ;Eickhout et al. 2008 ), apply yield projections from the FAO, which are presented for aggregated world regions level in Wirsenius et al. ( 2010 ). A comparison of these projections with historical growth figures indicates that in developed regions, the average annual yield growth rate is projected to be significantly lower than in the last five decades. For example, in North America and Oceania, the increase in beef yield is projected to decline from 1.0% year −1 in the period 1961-2005 to 0.2% year −1 from 1997/99-2030. Also on the global level, yield growth of beef and milk will be slower compared to the historical trend. Acceleration of yield improvements is projected to especially take place in sub-Saharan Africa. In this region, the yield growth rates of milk are projected to increase from −0.4% year −1  to 0.8% year −1 (1997/99-2030). In addition to the FAO projections, Wirsenius et al. ( 2010 ) defined an improved livestock production (ILP) scenario, assuming faster intensification of livestock production in low-and medium income regions as a result of increased competition for land and stricter policies related to land use and livestock production. In this scenario, more regions (e.g., Asia) will realize accelerated yield increases compared to the past. Also on the global level, yield growth will increase from 0.9% year −1 to 1.5% year −1 for beef and from 0.5% year −1 to 2.2% year −1 for milk. The scenarios from FAO and Wirsenius et al. again illustrate regional differences, which should be taken into account in the models.
To defi ne yield projections, it is again helpful to consider the potential role of different driving factors per region. In the historical assessment, it was found that the role of commercial livestock production for the domestic or export market is an important factor for explaining yield improvements. For modelling purposes, several scenarios could be defi ned for market development based on assumptions regarding the size and location of beef and milk consumption and production. In addition, the speed of yield improvements may be based on other driving factors like the possibilities for technological developments and the introduction of agri-environmental policies. Similar to crops, the technological improvement potential could be assessed through yield gap analysis. For livestock, however, no standardized methods exist to assess the yield gap ( ILRI ). One approach is similar to the conceptual framework for crops and is based on three groups of production factors; production defi ning (climate and animal genetic characteristics), production limiting (water and feed intake) and growth-reducing (diseases, pollutants) (van de Ven et al. 2003 ). This method is still new; the fi rst calculations of potential beef production were recently conducted by van der Linden et al. ( 2013 ).

FAO data
This study analyses a large amount of statistical data which is obtained from the FAOSTAT database. The quality of FAO data, however, can vary signifi cantly. When available, the FAO presents offi cial data , which means that the data are collected directly from the states. Yet, the data collection capacities and practices vary between countries and affect the reliability of the data. In addition, the concepts, defi nitions, coverage, and classifi cations used by the countries are not uniform and require harmonization to enable international comparison. When no offi cial data is available, the FAO gives fi gures from secondary semioffi cial or unoffi cial datasets or own estimations ( FAOSTAT ). With regard to crop yields used in this study, the amount of underlying data that is nonoffi cial data is limited and mainly restricted to Zambia and Zimbabwe. In the case of milk and beef yields, secondary and estimated data are more common and also presented on a regular basis for Brazil, China and India ( FAOSTAT ). Sometimes these data seem to be artifi cial as yields remain constant over one to fi ve decades (see for example yields from beef and milk production in Zambia and Zimbabwe, Fig. A9).
To get an impression of the reliability of FAO yield statistics, their consistency with USDA data was analyzed (Fig. A12, Appendix 9 ). In some cases, FAO and USDA diverge signifi cantly. For example, in Zambia, corn yield development between 1961 and 1982 is highly uncertain; fi gures from the FAO show an increasing yield trend, while the USDA data presents a downward trend. This has signifi cant impact on the interpretation of historical developments. The FAO data suggests that corn yield improvements started in the early 1970s with the introduction of fertilizer subsidies for smallholder farmers, while the USDA data implies that the yields were negatively affected by the new pricing and subsidy policies and only started to increase after the introduction of the fi rst new corn varieties in the late 1970s. Differences between the two statistical sources were also found for soybean in Zimbabwe, milk in Brazil and China, and beef in China and India. Not all data sets could be compared as the USDA database had no statistics on milk and beef production in Zambia and Zimbabwe and on sugarcane yields. The varying quality and reliability must be taken into account when interpreting the results. Still, the FAO database is the most complete source for yield fi gures currently available.

Yield indicators for cattle
With regard to cattle production, it is preferred to assess yield developments in terms of changes in feed conversion effi ciency (FCE, kg animal product per kg feed intake) instead of beef or milk production per animal. The main problem of using the production level per animal is that this fi gure does not always refl ect technological advancements. For example, an improved beef cattle production system may achieve faster weight gain and be able to reduce the cattle lifetime. As a result, the beef production per animal may remain constant or even reduce, while the total production can be increased. Also, de Wit et al. ( 2011 ) showed that beef yields in the Netherland decreased in the 1990s and 2000s because of the large share of dairy cows that are optimized for milk and not for meat production. In the present study, the historical data show that the average yield per animal has continued to increase in the main cattle producing countries. The rates at which these yields have increased, however, are likely to differ from improvement rates in feed conversion effi ciency.
Another reason why the use of the feed conversion effi ciency is preferred is the underlying idea of this study that yield improvements have an important role in making land available for biomass production without increasing overall land use. While crop yields are directly related to land use, fi gures of beef or milk production per animal give no indication of the related land use. As feed consumption can be linked more directly to land use, the feed conversion effi ciency would give a better insight in how developments in the cattle sector would infl uence land requirements.
Ideally, the FCE is measured over the lifetime of an animal because its value is not constant over time. To analyze historical developments in average FCE, a more simple but less accurate way is to calculate the feed conversion effi ciency by dividing the produced amount of beef or milk by the gross feed intake. This feed intake is based on estimated energy requirements and the amount of energy supplied by feed inputs, factors that are highly depending on the production system. Therefore, the examination of developments in feed conversion effi ciencies over time would require the allocation of animal populations and production quantities to the different production systems for at least several points in time, for example, building on previous work by Seré and Steinfeld ( 1996 ) and Bouwman et al. ( 2005 ). This was not feasible for the present study. As the carcass weight and annual milk production per animal are the best available data over a longer historical time period, these fi gures were used here to assess beef and milk yield developments. The same data were used by de Wit et al. ( 2011 ) and Wirsenius et al. ( 2010 ) to study livestock yield growth rates.

Yield projections and assessment of biomass potentials
This study investigated historical developments in yields and their drivers and provides suggestions for how potential studies can better account for yield developments and driving factors. The assessment focused on three types of driving factors: technological/management, economic, and institutional. Other factors, however, may also infl uence yield developments. Climate change, for example, may either have a positive or negative effect on yield growth depending on the location (see e.g., Jaggard et al. ( 2010 )). Also, several studies indicate that yield improvement rates of crops are related to the GDP level of a country (Hafner 2003 ;Powell and Rutten 2013 ). This correlation, however, does not necessarily mean that GDP itself is a driver of yield development. It is more likely that GDP is an indicator of other driving factors, such as market conditions and technology levels, which are included in the present study.
As shown in the section about yield projections, historical yield growth rates depend on the time frame considered. This is also seen when comparing the yield growth rates for France as calculated in this study and in de Wit et al. ( 2011 ). For wheat, for example, the present study found a growth rate of 4.3% year −1 (114 t ha −1 year −1 ) for the period 1961 to 1970 while de Wit et al. found a growth rate of 5.2% year −1 (136 t ha −1 year −1 ) for the period 1961 to 1969. Thus, the yield growth rates are highly sensitive to the timeframe applied, especially in the case of short time frames. The growth rates should thus be considered with great care and only be used as an indicator of the extent of yield growth or decline. Nevertheless, both studies show that the yield growth rates are very useful to assess the impact of driving factors on yield developments.
Although the historical assessment gives important insights into how different factors may infl uence future yield developments, it is not possible to predict future yield growth rates. The insights can thus only be used to assess how yields may develop under certain conditions. Particularly the application of endogenous factors and scenarios is useful to assess how yield developments change under different assumptions and how this affects biomass potentials. As mentioned in the section about yield projections, it is important to translate each scenario to conditions for meeting the projected yield developments. This can help identify (regional) strategies for increasing yields.
Finally, in addition to yields, there are also other factors that may affect biomass production potentials. For example, market developments and incentives could infl uence the balance between crop and livestock production on the one hand and biomass production on the other hand. Also, sustainability criteria could affect the area of land that is excluded from biomass production. An overview of more key factors is provided by Dornburg et al. ( 2010 ). Similar to the drivers for yield developments, it is important to make the assumptions regarding these factors explicit. Also, the application of scenarios could be useful.

Conclusions
Global, sustainable biomass production potentials of energy crops largely depend on the future availability of surplus agricultural lands made available through yield improvements in crop and livestock production. This study analyzed the pace and direction of historical yield developments between 1961 and 2010 in Australia, Brazil, China, India, the United States, Zambia, and Zimbabwe. Furthermore, it assessed the technological, economic, and institutional driving forces behind these developments and explored how the insights gained can help to improve the modeling of future yields.
This study showed that historical yield growth (especially of crops) has often followed a linear trend. Mainly, the average yield improvement rates for crops and milk were between 0.7% and 1.6% year −1 . For beef, the rates were lower (maximum of 0.8% year −1 in Australia; all relative to 2010). In all cases, yields and yield growth rates have fl uctuated to various degrees. Large fl uctuations were especially found for crops when driving factors changed strongly (e.g., extreme climate conditions in Australia). Also, in each case, the analysis revealed periods during which yields improved at a higher rate compared to the long-term average as well as periods during which yield growth rates were lower than this average. The periods of high yield growth, for example, 8.5% year −1 for soybean in Zimbabwe in the 1970s, show that relatively fast improvements can be attained in cases where the yield gap is still large. Such signifi cant improvements can especially be realized under favorable conditions with regard to economics and governance that stimulate improvements in agricultural technology and management. The future development of yields depends on how driving factors will change in each region.
The historical assessment shows that all three types of driving forces have infl uenced yield changes. The importance and the effect of each factor, however, is countryand even regional-specifi c. Overall, supporting agricultural policies have played an important role in increasing yields. Examples of successful policies are subsidies to stimulate adoption of new technologies, trade liberalization (resulting in increased demand for agricultural products which stimulated investments and innovations in the agricultural sector) and public investments in R&D. In some periods and countries, such policies were absent or eliminated (e.g., Australia in the 1960s and Zambia in the 1990s). As a result, yields stagnated or declined. Although agricultural policies led to yield increases in many cases, they failed to improve output-input ratios (i.e., unsuccessful to realize more effi cient use of resources like fertilizers). Some countries like the USA and (to a lesser extent) China were able to increase this productivity by implementing specifi c agri-environmental policies. Other countries adopted such policies as well, but the result largely depended on the success to enforce these policies (e.g., productivity stabilized in Australia, but no effect was seen in India). The importance of policies in steering yields was especially high for crops. With regard to yield improvements in cattle production, a key factor was the importance of commercial beef and milk production for the national or export market. But policy and market can be closely related: in many cases, trade liberalization created new markets, which stimulated investments and resulted in improved yields as demonstrated in, for example, Brazil. Current models that assess biomass potentials and impacts only take into account one or a limited number of endogenous factors infl uencing yields. Also, an explicit discussion of the assumptions behind yield projections is lacking, which hampers a comparison of yield projections between the models. Several suggestions are made to improve the models and thereby our understanding of potential future pathways for agricultural yield developments and for sustainable biomass production. First, scenarios based on regional assessment of key factors for yield development, as conducted in this study, could help to gain more insight in potential pathways and regionally differentiated effects. Second, to defi ne such scenarios, yield gap fi gures are an important indicator of possible technological progress and the potential rate of yield improvement. Also, different policy strategies should be included and tested in the scenarios. Finally, the assessment of important factors for yield development could help to make the underlying assumptions of yield projections more explicit. The implementation of these suggestions will help to identify policy options and preconditions for specifi c development pathways.

Acknowledgments
The authors thank Vassilis Daioglou (Utrecht University and PBL Netherlands Environmental Assessment Agency) for sharing yield data from the IMAGE model. David Laborde is thanked for sharing data from the MIRAGE model. This research was conducted within the research program "Knowledge Infrastructure for Sustainable Biomass", which is funded by the Dutch Ministries of 'Economic Affairs' and 'Infrastructure and the Environment'.

Confl ict of Interest
None declared.

Notes
1 Assuming continuation of current trends in diet and crop land area expansion. 2 Maximum attainable yield: the yield resulting from combining (1) the constraint-free potential agroclimatic yield with regard to temperature, radiation, and soil moisture conditions prevailing in the specifi c region and (2) reduction factors related to climate (e.g., pests and diseases), soil and terrain conditions, and assumptions regarding the management level ( GAEZ Global Agri-Ecological Zones ). Generally, it is assumed that there is a minimum yield gap where the actual yield level is equal to the economically attainable yield. Fischer et al. ( 2014 ) consider this economically attainable yield to be about 23% below the maximum attainable yield. Larger yield gaps are assumed to be 'economically exploitable yield gaps'. These caps could (largely) be closed with existing technologies (Fischer et al. 2014 ). 3 The high growth rate for soybean yields is also related to the drought induced yield drop in 1992. Without this outlier in the dataset, however, the improvement rate is still 12.2 % year −1 . Sugarcane yields were not considerably affected by the drought.
In the fi rst half of the 20th century, global agriculture had been affected by weather crises (droughts, fl oods, famines, plant, and animal diseases), the great depression in 1930, and two world wars (Tauger 2011 ;Hawkes et al. 2012 ). Later, after a decolonization process during the 1950s-1960s, the agricultural sector in former colonies was also underperforming (Hawkes et al. 2012 ). At the same time, agriculture had to fulfi ll an important role in supporting the industrialization process by providing cheap food to urban work force (Hawkes and Murphy 2010 ;Hawkes et al. 2012 ). As a result, during the 1940s-1970s, governments aimed to increase their production and continue the modernization and mechanization of agricultural systems which had started in the 19th century (Tauger 2011 ). To support and protect agricultural production and prices, states adopted agricultural protection policies. Since the 1950s, many developed countries imposed strong state controls on agriculture and trade through instruments like import tariffs, export subsidies and producer support (Hawkes and Murphy 2010 ;Smith et al. 2010 ;de Wit et al. 2011 ). These policy instruments guaranteed a minimum return to farmers and compensated for price differences between the internal and global market (Hawkes and Murphy 2010 ;de Wit et al. 2011 ). Developing and industrializing countries heavily taxed agricultural exports and protected producers from import competition (Hawkes et al. 2012 ). Rapid economic growth facilitated investments in agricultural R&D, which led to breeding of high-yielding crop varieties, increasing application of fertilizers and irrigation, and mechanical innovations (Piesse and Thirtle 2010 ;Tauger 2011 ;de Wit et al. 2011 ). This is referred to as the Green Revolution. By the 1970s, the Green revolution and state intervention resulted in overproduction in developed countries. Some less developed countries like China and Brazil were also able to increase agricultural production at a higher rate than population growth. But other developing countries (especially in Africa) lagged behind and suffered from underproduction (Hawkes et al. 2012 ). In the last 25 years, both developed and developing countries have begun to reform their agricultural policies, which has resulted in growing international trade of agricultural products.
Trade liberalization started with the fi rst General Agreement on Tariffs and Trade (GATT) in 1947. Although the trade of agricultural products was subject of international negotiations, it was typically excluded from multilateral trade agreements until 1990 (Hawkes and Murphy 2010 ;Hawkes et al. 2012 ). In 1995, the Uruguay Round Agreement on Agriculture (AoA) came into effect and imposed measures on signatory countries to open their agricultural markets (Hawkes and Murphy 2010 ;Hawkes et al. 2012 ). Another issue that remained unresolved in negotiations was the question how policies for sustainable development and environmental protection could be aligned with and integrated in trade regulations. Although environmental stewardship is a global issue, the interests in multilateral negotiations are diverse. While developed countries emphasize the need for an environmental reform of trade regulations, developing countries are mainly concerned with questions related to market access, dumping, and agricultural subsidies (Chambers 2002 ). As will be shown in the next sections, this diversity in interests is also refl ected in the varying degree of adoption and enforcement of agri-environmental policies in the selected countries.

Agricultural characteristics
Australia is a dry continent. Its climatic zones range from a tropical Northern region, through an arid interior, to a temperate Southern region (Pink 2012 ). The wet northern summer conditions allow beef cattle grazing and sugarcane production (east coast). The drier southern summer conditions favor wheat production, and grazing of sheep and dairy and beef cattle (Pink 2012 ). Rice is mainly grown in the Southeast of Australia (Pink 2012 ). About 10% of the agricultural area is cultivated (cropping and sown pastures and grasses) (Pink 2012 ;FAOSTAT ). The remaining area consists of permanent pastures and meadows for livestock grazing ( FAOSTAT ). The majority of the farms are engaged in either livestock farming or grain growing (Pink 2012 ).
The management levels are low by OECD standards. Until the early 1980s, about 3.5% of the cultivated land (arable land) was equipped for irrigation. From the late 1980s, this share increased to about 5.5% in recent years ( FAOSTAT ). Irrigation is mainly applied for vegetables and fruits, rice, and also sugarcane (Stringer and Anderson 2002 ;Pink 2012 ). Because low rainfall limits the returns on fertilizer expenditure, fertilizer use is relatively low (Stringer and Anderson 2002 ), see Figure 2 . The highest rates of fertilizer and pesticide are applied in horticulture (fruits and vegetables) (Stringer and Anderson 2002 ).
The average ruminant density on pastures is very low (see Fig. A10 in Appendix 9 ). Beef cattle are mainly held in Northern Australia, where production is extensive and the technology level low. In the South, production is more intensive. This is illustrated by higher stocking rates per hectare, improved pastures, and the use of fodder crops and animal health products (Pink 2012 ).
Dairy production mainly takes place in the south eastern high-rainfall coastal areas and is based on year round pasture grazing. Feedlot-based dairying is expanding, but is still uncommon (Pink 2012 ). Between 1980 and 2000, farmers switched to another dairy cattle breed (Doyle and Stockdale 2011 ).

Economic and institutional developments
Compared to other OECD countries, Australia has been more protective to industry for most of the 20th century. The country did not participate in the General Agreement on Tariffs and Trade (GATT) between 1947-1979. In the 1950s and 1960s, Australia ' s policies were focused on industrial protectionism, characterized by price support, and trade protection (e.g., import restrictions on manufacturing products). These policies isolated farmers from national and international market signals (Stringer and Anderson 2002 ) and resulted in indirect disincentives for agriculture  ). The subsidies and protection provided to the agricultural sector were limited and could not offset these disincentives  ).
From the 1970s, Australia has been reforming its trade policies  ). The past two decades have been a period of especially rapid total factor productivity (TFP) growth  ). One important factor explaining the increase is the openness of the Australian economy to trade and investment (Fischer et al. 2014 ).
In 2002, Stringer and Anderson ( 2002 ) mentions that "four-fi fths of the Australian agricultural production is exported". The main export markets are the United States and Asia (Stringer and Anderson 2002 ).
Until the 1990s, agricultural policies were led by socioeconomic objectives. Since the 1990s, more emphasis has been put on sustainable agricultural development (Stringer and Anderson 2002 ). Current agricultural policies aim to improve market responsiveness, and encourage sustainable agricultural practice (i.e., approaches that combine economic, environmental, and social aspects). Projects also focus on food quality. Agri-environmental policies concentrate on water, soil erosion, salinity, and biodiversity loss (Stringer and Anderson 2002 ). Measures to achieve this include research, education, voluntary adoption of best practice, and development of guidelines in collaborations between governments, NGOs, industries, and communities. Over time, regulatory approaches have been complemented, or even substituted, with market oriented mechanisms like for example the polluter pays principle. The basis of these measures is that prices refl ect social costs and benefi ts, that is, positive and negative externalities are taken into account (Stringer and Anderson 2002 ).

Yield developments
In Australia, a small average change in wheat and rice yields between 1961 and 1980 was followed by a signifi cant increase in the 1980s (Fig. 3 and Table 1 ). As this coincides with the participation in the General Agreement on Tariffs and Trade (GATT) after 1979, it is likely that the improvements are a result of the trade policy reforms which opened the international market  ). Notably, the input use levels remained stable. Thus, it seems that the reforms motivated farmers to use their resources more effi ciently, which resulted in higher production levels and improved productivity in the early 1980s (Fig. A1 ). Agri-environmental policies were fi rst implemented in the 1990s and aimed to improve market responsiveness and encourage sustainable agricultural practice (Stringer and Anderson 2002 ). Since their introduction, agricultural production has further increased. But, fertilizer use and irrigation levels increased as well and the productivity of the agricultural sector did not improve compared to the 1980s (Fig. A1 ). Also, wheat and rice yields stagnated in the 1990s. Sugarcane yields have been relatively steady and did not signifi cantly improve in the 1980s. A considerable drop in yield in the period 1990-92 was followed by a peak in the second half of the 1990s and another plunge in 2001-02 due to drought. Then, sugarcane yields seem to have stabilized again around previous levels.
Beef and milk yields have almost continuously improved, except for a period of stagnation in the 1970s and early 1980s ( Fig. A9 and Table 1 ). In this period of market reforms, also the production and export of beef and milk temporarily declined ( FAOSTAT ). Between 1980 and 2000, beef and milk yields improved while their export markets grew, feed crop consumption increased enormously (Fig. A1 ), and a shift in dairy cattle breed from British breeds to Holstein-Friesian animals was made (Trewing 2004 ;Doyle and Stockdale 2011 ). Although the majority of cattle production is still extensive, the rise in feed crop consumption is likely to be related to the intensifi cation of beef production in South Australia and the expansion of feedlot-based dairying (Pink 2012 ). The above fi ndings suggest that the development of export markets has been an important driver for changes in the production systems of beef and milk, and all these factors together have contributed to yield improvements. Due to the massive increase in feed crop use, however, the output-input ratio of the livestock sector has decreased signifi cantly. The decline has slowed down since the mid-1990s, which may be related to the introduction of agri-environmental policies.

Agricultural characteristics
Agriculture in Brazil is characterized by concentrated land ownership; medium-and large-scale commercial farms contribute the bulk of agricultural output (Anderson and Valdés 2009 ). Crop and livestock are combined in mixed farming systems (Dixon et al. 2001 ). The majority of cattle is held in extensive mixed (rain-fed) production systems (Wint and Robinson 2007 ). These are found in the wooded and open savannah areas (the Cerrados) in the Central West of Brazil, and also in the Southeast (Dixon et al. 2001 ;Cederberg et al. 2009 ). Extensive ranching is the primary activity, but cultivation of soya and corn is increasing. In addition to this farming type, i ntensive mixed farming takes place in Eastern and Central Brazil. This system produces most of the sugarcane, which is mainly cultivated in the Central South of Brazil. D ryland mixed farming (which is mainly semisubsistence farming) is the major system in northeastern Brazil (Dixon et al. 2001 ).
A signifi cant expansion of the agricultural sector in the last decades was accompanied by a considerable increase in deforestation, replacement of native vegetation, and biodiversity loss (Martinelli et al. 2010 ;Ferreira et al. 2012 ). Also, the use of fertilizers and other inputs, and the ruminant density on meadows and pastures have risen    (Figs 2 and A10 ). The ruminant density in Brazil is high compared to the other selected countries, except to India. The largest share of the cattle population is being held for beef production; about 10% are dairy cows (Cederberg et al. 2009 ;FAOSTAT ). Beef and milk production are mainly based on extensive systems, in which cattle grazes on pastures all year round (Cederberg et al. 2009 ). In the emerging semiextensive systems, herds also receive supplemental feed from crops and various concentrates. Feedlot-based, intensive systems are still rather uncommon (Cederberg et al. 2009 ).

Economic and institutional developments
Prior to 1950, the Brazilian market was concentrated around the export of food and raw materials, and the import of industrial products (Baer 1972 ). Between 1950 and the mid-1970s, the focus shifted to national industrialization, and policies aimed at replacing foreign imports with domestic production (Carvalho 1991 ). This is called Import Substitution Industrialization (ISI) (Carmo Oliveira 1986 ). To protect the industry, wage rates were kept low by restrained food prices (Carvalho 1991 ). In order to realize low food prices, Brazilian agriculture was heavily and increasingly taxed (Carmo Oliveira 1986 ). Levies consisted partly of direct export taxes, but were dominated by indirect taxation resulting from industrial protection policies (Anderson and Valdés 2009 ). Due to the industrial protection, also input prices increased. Therefore, the government provided credit and fertilizer subsidy to promote the use of fertilizer and other inputs (Carvalho 1991 ).
Trade liberalization started in the 1980s, and continued to the mid-1990 ' s (Anderson and Valdés 2009 ). Reforms included the removal of import and export restrictions, and the redistribution of resources from import-competing to export-oriented sectors (Anderson and Valdés 2009 ). This transformation took also place in the agricultural sector (Anderson and Valdés 2009 ). Today, major exports of agricultural products like soybean, sugar, beef, and ethanol contribute to Brazil ' s positive balance of trade. Thus, the agricultural sector plays an important role in the economic development of Brazil (Martinelli et al. 2010 ;Ferreira et al. 2012 ).
During the last years of industrial protectionism, the oil crisis in 1973 prompted the Brazilian government to phase out fossil fuels. The ProÁlcool program was launched to promote the sugarcane industry and bio-ethanol production (Stattman et al. 2013 ). Blending ethanol to fossil fuel was already introduced in 1931, but the ProÁlcool program brought about a major increase in ethanol consumption and production (Stattman et al. 2013 ). Today, most of the ethanol produced is still intended for the domestic market. In 2009, almost 14% of the production was exported (Lamers et al. 2011 ). To improve Brazil ' s energy diversity and independence, a second biofuel program was implemented in 2004: the National Program of Production and Use of Biodiesel (PNPB). Due to the abundance of soy and the search for new soy markets, biodiesel production is largely based on the conversion of soybean oil (Stattman et al. 2013 ).
The need to control deforestation was already recognized in the 1920s. The fi rst Forestry Code, which dates from 1934, regulated the conservation of forests on private land (Banerjee et al. 2009 ). In 1965, a second code expanded the land dedicated to preservation from forests to other sensitive areas. Also, it created conservation areas outside the private rural properties (Banerjee et al. 2009 ). Due to economic priorities, however, enforcement of the codes was weak. Between 1974 and 1987, when the focus shifted from protectionism to trade liberalization, the government promoted livestock production, forestry, and mining in the Brazilian Amazon (Banerjee et al. 2009 ). The markets for Brazilian beef have been growing since the 1970s and led to considerable expansion of extensive cattle ranching on cleared forest land in this region (Bowman et al. 2012 ). In addition, the more recent expansion of soy production on previous pastures causes further expansion of cattle ranching into the Amazon (Bowman et al. 2012 ). Policies and other initiatives that aim to intensify cattle production are in early stages yet (Bowman et al. 2012 ).

Yield developments
In Brazil, the developments in the yield of corn and soybeans are comparable. From the 1960s to 1980s, yield growth rates were moderate and highest in the 1970s (Fig. 3 and Table 1 ). In the 1990s, high improvement rates of more than 3.5% year −1 were attained (relative to 1991). Fertilizer use increased substantially in the late 1960s and the 1970s, declined and stagnated in the 1980s and increased again in the 1990s (Fig. 2 ). Thus, it appears that fertilizer subsidies provided during industrial protection (1950s-70s) (Carvalho 1991 ) led to an increase of fertilizer use and of yields, especially in the 1970s. The economic reforms in the 1980s (Anderson and Valdés 2009 ) temporarily hindered agricultural development, but the opening of agricultural export markets stimulated further improvements in the 1990s. Notably, after agricultural productivity declined in the 1960s and 1970s, the output-input ratio has remained fairly constant from the 1980s until 2002 (Fig. A2 ). Despite the introduction of the biodiesel program in 2004 (Stattman et al. 2013 ), soybean yields stagnated in this decennium. Yet, the share of soybeans used for biodiesel production has been rather small in the fi rst years (0.5% in 2006 compared to 12% in 2010) (Barros 2013 ;FAOSTAT ). The development of sugarcane yields is similar to the trend for corn, but the fi rst period of major growth is found between 1975 and 1985. This is clearly related to the introduction of the PróAlcool program in the early 1970s (Stattman et al. 2013 ). Tractor use and irrigation also grew signifi cantly after 1973, thus are probably related to the rise of sugarcane production.
Considering beef and cow milk production, no significant yield improvements were attained from 1961 until the 1980s (Fig. A9 and Table 1 ). This period of relatively stable yields was interrupted by a few years of decline in the mid-1970s. Major yield increases were only attained in the early 1990s. This suggests that the promotion and expansion of cattle ranching during the period of liberalization (Banerjee et al. 2009 ;Bowman et al. 2012 ) did not directly stimulate yield improvements. Only when the export markets were fully opened in the early 1990s, yields signifi cantly improved. For milk, this growth continued in the late 1990s and 2000s, but at a lower rate. The initial increase in beef yields was fi rst followed by a decline in the late 1990s, before yields increased again in the 2000s. As the production of beef and milk has grown signifi cantly faster than the use of feed crops and the area of pasture land, the output-input ratio of the livestock sector has continuously increased over the past fi ve decades, especially since the mid-1990s (Fig. A2 ).

Agricultural characteristics
Agriculture in China is characterized by large environmental diversity, and large diversity in agricultural products.  Promotion of livestock production, mining and forestry in Amazon Trade liberalization, from import competing to export oriented PNPB The Qinling mountains divide China into water-defi cit (North, West) and water-surplus regions (South, Northeast) (Veeck et al. 2011 ). There are four major farming systems (Dixon et al. 2001 ). Lowland rice production is found in humid and moist sub humid areas in South and Central East China. Rice production is rain-fed, with supplementary irrigation where available. Important livelihoods are, besides rice, subsidiary crops like corn and soybeans, livestock, and off-farm work. Upland Intensive Mixed Farming is found in upland and hill areas with humid and subhumid climate (South East and North China). A signifi cant area, mainly rice, is irrigated. Livestock contributes to draught power, meat, income and savings. Also off-farm work is an important income source. Temperate mixed farming is found in moist and dry subhumid areas in Central Northern China. The major crops are wheat and corn. Livestock is also an important livelihood.
Pastoral farming is located in semiarid and arid temperate climates in Western China. Pastoralism is based on extensive grazing of mixed herds (camels, cattle, sheep and goats) on native pasture. In local suitable areas, farmers apply irrigated crop production (e.g., cotton, barley, wheat). Characteristic for the agricultural sector of China are the majority of small-scale farms and the application of multiple cropping systems, that is, the production of more than one crop per year on the same land (Veeck et al. 2011 ). China aims to be largely self-suffi cient in grain production. Because of its large population, land availability for agricultural production is an important issue (Veeck et al. 2011 ). Economic growth has caused a major increase in population and demand for housing, transport, and industry in the Eastern coastal area. Much land in this region, however, is fertile and highly productive agricultural land which is lost due to the urbanization process (Karplus and Deng, 2008a ;Veeck et al. 2011 ). Due to this pressure, even marginal lands (e.g., with very limited precipitation or extreme slope) are cultivated (Veeck et al. 2011 ).
Without irrigation, the dry areas are of marginal use for intensive agriculture. Water shortages hamper the improvement of agricultural production in these regions. Therefore, irrigation has expanded at a high rate (Fig. 2 ). This, however, is causing severe water shortages as water consumption outpaces replacement through precipitation; there are major concerns that groundwater reserves are being depleted, especially in arid areas in Eastern and Western China (Karplus and Deng, 2008a ;Veeck et al. 2011 ). Fertilizer use has increased dramatically in the past fi ve decades (Fig. 2 ). Fertilizer use is especially high in lowland rice and in temperate mixed farming systems (Dixon et al. 2001 ). The overuse of fertilizers is associated with land degradation, air pollution, and eutrophication of water sources (Veeck et al. 2011 ).

Economic and institutional developments
The period from 1949-1976 in China is called the Maoist era. During this era, the Chinese Communist Party (CCP) was in power. Governance was characterized by a strong inward orientation and self-imposed isolation (Veeck et al. 2011 ). Also, policies were relatively homogenous for the nation as a whole, and the use of capital and resources was regulated centrally (Veeck et al. 2011 ). In this era, farming took place in large farming communes (Tauger 2011 ). These communes were diffi cult to manage. In 1959, this caused a collapse of farm production and a huge famine. In order to solve the problems, agricultural reforms were introduced which included the reparation and construction of irrigation systems and the distribution of high-yield seeds (Karplus and Deng, 2008a ;Tauger 2011 ). Programs to develop improved crop varieties had already started in the early 1950s. The communes were broken up when the market reforms started in 1978. Since then, individual farmers have been leasing land from the local authorities. The resulting diversifi cation of crop production and farmers' activities, income, and education level became visible in the second half of the 1990s (Veeck et al. 2011 ).
Since the end of the Maoist era, the CCP has still been in charge. The central government continues to play an important role in planning and guiding the direction of development (e.g., economic decision making), but the role for local governments is increasing (Veeck et al. 2011 ). Also, from 1978, the focus of the market shifted from import substitution industrialization toward export-oriented development strategies (Anderson and Martin 2009 ). This resulted in an export-led industrial growth, and also a restructuring of the economy away from agriculture and heavy industry toward light manufacturing and service activities (Anderson and Martin 2009 ;Veeck et al. 2011 ). The taxation of agricultural exports has been reduced, but the protection of import-competing agriculture, especially of rice, has been increased (Anderson and Martin 2009 ). Because of the importance of the agricultural sector, the government increased investments in agricultural R&D and started to fund research in biotechnology. This support was continued in the following decades (Karplus and Deng, 2008b ).
Environmental protection laws were fi rst introduced in the late 1980s. These laws aimed to prevent the loss of high-productivity cropland caused by the expansion of urban and industrial areas in Eastern China (Veeck et al. 2011 ). Also, the increased awareness about environmental problems and the need for more effi cient agriculture led to the implementation of the Comprehensive Agricultural Development (CAD) program in 1988 (Veeck et al. 2011 ). The CAD program was introduced because of the low productivity of a large share of arable land and increasing grain imports. The program aimed to enhance the quality of agricultural land through better land management, including improved fertility and drainage, balanced use of inorganic fertilizer, irrigation and water storage and conservation. The CAD still exists (Veeck et al. 2011 ).

Yield developements
Since the agricultural reforms in the early 1960s (Karplus and Deng, 2008a ;Tauger 2011 ), irrigation, fertilizer use, and mechanization have been increasing almost continuously (Fig. 2 ). In addition, fast yield growth is found for all crops (corn, rice and wheat), Table 1 . On average, the average yield growth rate was the highest for wheat (1.8% year −1 relative to 2010), but the highest absolute improvement was achieved by corn and rice (93 kg ha −1 year −2 ). These considerable gains can mainly be attributed to the introduction of new technology, which was realized by signifi cant public investments in infrastructure and research (Karplus and Deng, 2008a ). In the 1990s, however, yield improvements dropped. This may refl ect the diversifi cation of farmers' crop production after the economic reforms (Veeck et al. 2011 ). In the following decade, the growth rates of corn and wheat yields increased again, while rice yield improvements continued to slow down.
The yield growth rates for beef and milk have been lower than for crops. This may be explained by the major importance of cattle for delivering draught power, as agricultural production is still labor-intensive in China (Fig. 2 ). The market reforms may have resulted in some improvements in cattle production; the beef and milk yield shortly increased around the 1980s. Afterwards, beef yields stabilized again. Milk yields also stagnated for some years, but have been increasing at a rate of 3.5% year −1 since the late 1990s (relative to 2001, Fig. A9 and Table 1 ). At the same time, the consumption of milk and dairy products Livestock sector Index, 1961 = 100 Agricultural sector by urban residents soared, which was caused by China ' s growing prosperity (Bao 2011 ).
After the output-input ratio of the agricultural sector had declined rapidly in the 1960s and 1970s, it stabilized in the 1980s and improved gradually in the 1990s (Fig. A3 ). It is likely that both market liberalization and agrienvironmental policies, which aim to improve agricultural land quality (Veeck et al. 2011 ), have contributed to this reversal of the downward trend in productivity.

Agricultural characteristics
India has two major farming systems: rice-wheat and rain-fed mixed. The rice-wheat system in Northern India is characterized by wetland rice production in summer (monsoon season), and irrigated wheat production in winter (cool, dry season). A signifi cant amount of livestock is held in this system, where bovines produce draft power, milk and manure for composting (Dixon et al. 2001 ).
The rain-fed mixed system occupies the largest area in India (Central and Southern India). It is mainly rain dependent, but according to Dixon et al. ( 2001 ), about 16 percent of the area cultivated under this system was equipped with simple, small-scale irrigation techniques around 2000. Infrastructure and market access are poor, and agricultural activities are oriented toward subsistence. The main livelihoods are cereals, legumes, fodder crops, livestock and off-farm activities (Dixon et al. 2001 ).
The input-intensity of the agricultural sector has increased substantially since 1961 (Fig. 2 ). Yet, as the increase in input use outpaced the growth in total production, the output-input ratio has declined seriously (Fig. A4 ). This has caused major environmental issues. Large-scale irrigation in Northern India has infl icted soil salinization and groundwater depletion (Shah 2012 ). Also, groundwater is polluted due to intensive use of fertilizers and rudimentary processing of livestock wastes. In addition, large livestock populations cause soil degradation through the conversion of natural vegetation (Shah 2012 ). In Southern India, soil erosion is the main problem. The vegetative cover and organic matter content of soils are low. Yet, farmers continue to cultivate crops on marginal lands to meet their basic needs (Shah 2012 ).

Economic and institutional developments
India is a former colony of the United Kingdom and gained independence in 1947. In order to prevent famines, and to ensure affordable prices for basic foods, the Indian government has been intervening in the food market since its independence in 1947. In the public distribution system, which was established in 1958 and is still present, basic foods are sold at subsidized prices (Gulati and Pursell 2009 ).
In response to droughts and famines in 1965-66, policies aimed at food grain self-suffi ciency and agricultural imports began to be replaced by domestic production (Gulati and Pursell 2009 ;Tauger 2011 ). Green revolution technologies played an important role, as the government implemented many programs to modernize agriculture at a high speed. This included the development and planting of high-yielding wheat and rice varieties and large subsidies for electricity and fertilizers (Gulati and Pursell 2009 ;Tauger 2011 ). According to Dixon et al. ( 2001 ), however, agricultural development during India ' s Green Revolution did mainly take place in 10 percent of India ' s districts which had adequate local infrastructure for water management, transport and electricity (for tubewells). In the 1970s and early 1980s, the import of edible oils expanded signifi cantly. This led to policies which aimed to decline these imports and substitute them with domestically produced oils (Gulati and Pursell 2009 ). Import substitution was abandoned in the 1990s and the focus shifted to an export oriented economy. Trade policies were reformed through the structural adjustment program (SAP), which was introduced 1991 (Mukherjee and Chakraborty 2012 ).
A signifi cant number of environmental policies exist that aim to control the use and management of natural resources. Enforcement of these regulations, however, is weak (Shah 2012 ).

Yield developments
In India, crop yields have almost continuously grown in the last fi ve decades, but at different rates. The highest rates are found for wheat and rice in the period 1961-1990 (Table 1 ). Explanations for these achievements can be found in the major investments by the government in the modernization of agriculture (in Northern India), the development and adoption of high-yielding rice and wheat varieties and agricultural subsidies (Dixon et al. 2001 ;Gulati and Pursell 2009 ;Tauger 2011 ). After 1990, the absolute and relative yield growth of these crops has decreased. Soybean yields started to increase in 1972. This coincides with the introduction of the import substitution policy for edible oils (Gulati and Pursell 2009 ). Since 1961, sugarcane yields have increased at a moderate rate until 1999. After a decline in 2000-2004, yields returned to the level of the mid-1990s. For all crops, there is no clear relation between market reforms and yield developments in the 1990s. Although the linear regression data suggest that yield growth of rice, wheat and soybeans slowed down in this decade, the graphs in Figure 3 do not show a signifi cant deviation from an earlier trend which can be linked to the liberalization process.
Regarding cattle product yields, there is a signifi cant difference in developments between milk and beef. Cow milk yields have increased all fi ve decades, and growth rates increased as well. Similar to China, this is likely to be related to increased milk consumption ( FAOSTAT ). Beef yields are low compared to the other six countries. This is likely to be related to the protected status of cows in Hinduism, the major religion in India. Beef yields have been rather constant and only increased in the 1970s and 1980s. This temporary improvement may be explained by reduced need for draft power due to the mechanization process (Fig. 2 ). Mechanization, however, has continued in 1990s, but this is not refl ected in further improvement of beef yields.
Due to the enormous increase in inputs, the agricultural productivity has continuously decreased between 1961 and 2000 (Fig. A4 ). Although the decline has slowed down, the lack of productivity improvements confi rms the weak enforcement of agri-environmental policies in India (Shah 2012 ).

Agricultural characteristics
In the United States, agricultural production is mainly concentrated in the Pacifi c and Central (Midwestern) regions and the Southern plains (Alston et al., 2010a ). Over the past decades, the total number of farms in the United States has declined but the number of large scale farms has increased. Still, large-scale farms are a minority of all US farms, but they produce more than two-third of agricultural output (Anderson and Valdés 2009 ;Alston et al., 2010a ). The production practices depend on the farm size and the natural resource base (e.g., soil moisture and fertility). For example, the major practice in the corn belt (Midwestern USA) is dryland farming. In the Central   (Carpentier and Ervin 2002 ). The agricultural production of the last fi ve decades can be characterized by intensive management, especially with regard to mechanization (Fig. 2 ). It seems, however, that the intensity of input uses has stabilized since the 1970s and 1980s. Also, the ruminant density on pastures has declined (Fig. A10). In the livestock sector, dairy production is mainly confi nement based (McCormick 2011 ). Pasture use, however, has grown since the early 1990s (McCormick 2011 ). In pasture based production systems, dairy cows may be at pasture during parts of the year. At the same time, and in winter months, they receive stored forage along with varied levels of supplemental concentrates throughout the year (McCormick 2011 ). Beef production is mainly characterized by cow-calf herds on pasture and (winter) hay (Pelletier et al. 2010 ). Beef cattle are fi nished in feedlots where they receive a mixed, high concentrate feed ration. Less than 1% of beef cattle are fi nished in pastures (Pelletier et al. 2010 ). On pastures, no housing is provided for cow-calf herds. Hormone implants are employed in the feedlot stage. Calves can also be sent to feedlots directly. Pelletier et al. ( 2010 ) mentions that this is common practice in the US Upper Midwest.
The high intensity of agricultural management in the United States has led to a wide range of environmental issues. The most important problems are soil erosion (i.e., loss of the fertility and water-holding capacity of the soil) and contamination of water sources by agricultural chemicals and livestock manure (Carpentier and Ervin 2002 ).

Economic and institutional developments
From their introduction in the 1930s, agricultural policies in the United States have been differing signifi cantly in composition from the EU and other OECD countries. The focus of agricultural policies has been on providing food aid and nutrition assistance. Assistance to farmers in the form of commodity support programs is placed second (Silvis and van Rijswick 2002 ). These commodity programs consisted of price support and direct income payments (Silvis and van Rijswick 2002 ;Blandford and Boisvert 2006 ). To limit payments by the government, crop programs, such as corn and wheat, placed limits on production. For other commodities, like milk, import restrictions were applied (Blandford and Boisvert 2006 ). Market liberalization and multilateral trade agreements have changed the programs for farm support. Since 1985, income support has shifted to payments that are decoupled from prices and production. Also, production limitations have been replaced by more planting fl exibility, enabling farmers to make market-based decisions (Blandford and Boisvert 2006 ).
The United States have been dominant in agricultural R&D expenditures (Alston et al., 2010b ). For example, Figure A5 shows that American public spending in 2000 was twice as high compared to the investments made by China. In addition, agricultural R&D in the United States has been funded extensively by the private sector. The public and private sector contribute both about half of the total investments (Alston et al., 2010b ). In other countries, especially the developing countries, the share of private spending has been much smaller (Fig. A6 ). In addition, innovations in agricultural technology have been stimulated by intellectual property rights. Until the 1970s, however, this protection excluded inventions related to living organisms like plants and animals (Alston et al., 2010c ). This changed in the 1980s and the legalization of patents on life forms cleared the way for biotechnology to rapidly expand (Alston et al., 2010c ).
Agri-environmental policies in the US have been implemented from about 1970 (Carpentier and Ervin 2002 ). Traditionally, broad programs were implemented in each state. More recently, individual state and local programs  have emerged (Carpentier and Ervin 2002 ). The majority of agri-environmental policies have been voluntary-payment programs (Carpentier and Ervin 2002 ). The application of regulatory approaches which defi ne input or performance standards for agriculture was limited by two factors. First, agriculture in the United States is characterized by a high variety in production practices and local circumstances (ecosystems, quality and sensitivity of resources). Regulation of this diversity in operations is technically diffi cult and expensive. Second, a strong agricultural lobby exists in the United States, and political infl uence on environmental protection has been modest (Carpentier and Ervin 2002 ). As a result, regulatory programs and environmental standards for agriculture have only been introduced since the second half of the 1990 ' s. These regulatory programs focus on the quantity and quality of production inputs (especially water quality in livestock production, and pesticide use) (Carpentier and Ervin 2002 ). The most important environmental policies in the United States are the Conservation Reserve Program (CRP, farm bill 1985) and the Environmental Quality Incentives Program (EQIP, farm bill 1996). The CRP is a voluntary program that provides payments to farmers who apply conservation practices on environmentally sensitive lands (Silvis and van Rijswick 2002 ). The EQIP provides fi nancial and technical assistance to farmers to improve and protect the environmental quality of their properties (e.g., soil and water) (Silvis and van Rijswick 2002 ).

Yield developments
In the United States, the crop yield growth trends have been positive for most of the period 1961-2010 (Fig. 3 ). It is very likely that the substantial investments in agricultural R&D (Alston et al., 2010b ) have played an important role in achieving these improvements. From the mid-1980s, growth of corn and wheat yields slowed down for about a decade. Probably, this deceleration is related to the reforms of trade and farmer support policies in the same period (Blandford and Boisvert 2006 ). Afterwards, however, absolute growth reached a record high in the 1990s (Table 1 ). For soybeans, growth accelerated in the 2000s. In addition to the effect of trade liberalization, these signifi cant increases in yield growth may be attributed to the rise of biotechnology since the 1980s (Alston et al., 2010c ). Improvements in technology and management have also driven yield growth in beef and cow milk production. This technological progress is likely to be stimulated by investments in R&D and growing domestic milk consumption (Alston et al., 2010b ;FAOSTAT ). Although absolute and relative yield growth of beef was highest in the 1960s, yields have also been increasing considerably since the mid-1980s after a period of stagnation in the 1970s. Cow milk yields have almost continuously increased and while the relative growth rate has been fairly constant, the absolute growth accelerated in the 1980s and 1990s.
Regarding agricultural management, tractor use peaked in 1966 and fertilizer use reached the highest level in 1980 (Fig. 2 ). Also, the output-input ratios of the agricultural and livestock have been improving since the 1970s (Fig. A7 ). As agri-environmental programs were introduced in the same decade, the developments in input use and agricultural productivity may well be related to these policies.

Agricultural characteristics
The agricultural sector in Zambia exists of a small number of large-scale commercial farmers who have good access to input and output markets, a few medium-scale commercial farmers for whom market access is diffi cult, and a majority of smallholders who are often engaged in subsistence farming (Howard and Mungoma 1996 ;Bonaglia 2009 ). The large-scale farmers produce and sell wheat, soybean, coffee, milk and other livestock products. Corn, however, dominates the agricultural sector and is mainly produced by the smallholders and medium-scale commercial farmers (Howard and Mungoma 1996 ).
The major farming systems in Zambia are maize mixed (Central and East Zambia) and cereal-root crop mixed (West Zambia) (Dixon et al. 2001 ). Maize mixed systems are found in plateau and highland areas with a dry subhumid to moist subhumid climate. Besides corn, principal livelihoods are tobacco, cotton, cattle, goats, poultry, and off-farm work.  Figure A6 . Public share in total agricultural R&D spending in 2000 (Alston et al., 2010b ).
Driving Factors for Yield Developments S. Gerssen-Gondelach et al.
Cattle are kept for ploughing, milk, manure, but also for savings (Dixon et al. 2001 ). Cereal-root crop mixed systems are situated in regions of lower altitude and higher temperatures. The number of livestock per household is higher compared to the maize mixed system. The major sources of income are corn, sorghum, millet, cassava, yams, legumes, and cattle (Dixon et al. 2001 ).
In Zambia and other South African countries, the major environmental problem related to agriculture is declining soil fertility (Dixon et al. 2001 ). Soil degradation is caused by inappropriate management practices such as continuous cropping and overgrazing (Dixon et al. 2001 ). Average agricultural input levels in Zambia are low (Fig. 2 ).

Economic and institutional developments
Zambia is a former colony of Britain (the region was named Northern Rhodesia) and gained independence in 1964. After independence, the economy heavily depended on copper exports and many people lived in the urban mining areas. To ensure food supply to these areas, the new government aimed to increase national corn production. In the colonial period, however, commercial corn production had mainly relied on large-scale European farmers. The new objective was to enhance the participation of smallholders in the commercial corn market (Howard and Mungoma 1996 ). The agricultural intervention system of price controls and subsidies, which also dated from the colonial period, was maintained. But, new pricing policies favored smallholders in remote areas over commercial farmers with good market access (Howard and Mungoma 1996 ;Pletcher 2000 ). In the early 1970s, agricultural policies were expanded with fertilizer subsidies, which mainly benefi tted the corn sector (Howard and Mungoma 1996 ;Pletcher 2000 ). Both corn seed and fertilizer were made accessible to smallholder farmers in   In addition, farmers could sell their corn to these depots (Howard and Mungoma 1996 ). In the meantime, the Zambian government had invested in a corn breeding program which resulted in the release of twelve new varieties between 1977 and 1994. The program was started in the early 1960s to reduce the import of crop varieties from Zimbabwe, on which European farmers had relied during the colonial period (Howard and Mungoma 1996 ). Between 1973 and 1991, Zambia had been governed by single party rule. This period coincided with an economic crisis in the late 1970s and 1980s due to a collapse of the copper price in 1975 and poorly managed governmental interventions in the market (Pletcher 2000 ). Although attempts were made to reform (agricultural) policies in the 1980s, economic liberalization only started when a new government came to power in 1991 (Pletcher 2000 ;Jayne et al. 2003 ). Through liberalization, the corn market was fully privatized. But, intervention in the input markets for fertilizer and credit remained (Pletcher 2000 ). Fertilizer price subsidies had been eliminated in 1988, which resulted in high input costs for corn. Therefore, smallholders reduced their use of fertilizer and hybrid corn varieties and returned to the cultivation of traditional corn varieties and subsistence crops like sorghum (Howard and Mungoma 1996 ;Dixon et al. 2001 ). The government then decided to continue fertilizer distribution on loan, but this undermined the ability of the private market to distribute fertilizer commercially (Jayne et al. 2003 ). Also, underinvestment in infrastructure and other public goods had made the purchase of fertilizer unprofi table to many farmers (Jayne et al. 2003 ). In response to the reduced fertilizer use (Fig. 2 ) and corn production in the 1990s, a new policy for fertilizer distribution and subsidy (the Fertilizer Support Program) was implemented in 2002 (Djurfeldt et al. 2011 ).
Efforts to control soil erosion started in the mid-1980s, driven by the spreading problem of land degradation and the economic reforms in late 1980s and early 1990s. At fi rst, commercial farmers adopted conservation farming technologies to improve the profi tability of mechanized corn production (Haggblade and Tembo 2003 ). In 1995, appropriate technologies for smallholders were introduced as well. The development and promotion of the technologies was collectively conducted by farmer organizations, private companies, NGOs and the government (Haggblade and Tembo 2003 ).

Yield developments
After corn yields declined and fertilizer use increased slowly in the 1960s, the introduction of fertilizer subsidies (Howard and Mungoma 1996 ;Pletcher 2000 ) caused these levels to increase signifi cantly in the 1970s (Fig. 2 , Table 1 ). The fact that agricultural policies were mainly focused on corn production (Howard and Mungoma 1996 ;Pletcher 2000 ) is clearly refl ected in the high yield improvement rate of 4.5% year −1 for corn compared to 1.2 % year −1 for soybeans and 0.1 % year −1 for sugarcane between 1971 and 1990 (relative to 1971, 1973 for soy). After the elimination of fertilizer subsidy in 1988 due to economic liberalization (Howard and Mungoma 1996 ;Pletcher 2000 ), fertilizer use and corn yields declined in the 1990s. It appears that commercial farmers benefi tted from the economic reforms, as irrigation levels increased considerably and soybean yields improved at a very high rate of almost 20 % year −1 in the 1990s 3 (relative to 1991). Sugarcane yields increased at 0.7 % year −1 in the same decennium. In addition, Figure A8 shows that the output-input ratio of Zambia ' s agricultural sector improved in the late 1980s and 1990s. Besides reduced fertilizer use, these advances in productivity may also be the result of the introduction of conservation farming technologies (Haggblade and Tembo 2003 ). After the adoption of the Fertilizer Support Program in 2002 (Djurfeldt et al. 2011 ), however, fertilizer use increased again, the area equipped for irrigation stabilized (Fig. 2 ) and overall agricultural productivity dropped. Corn yields rose again and sugarcane yields stabilized. The effect on soybean yields is unclear; after a steep decline in 2001 due to drought, yields recovered and returned to levels comparable to the 1990s.
The FAO data show constant milk yields from 1961 until 2010. Beef yields have also been relatively stable, except for a decline of 1.8% year −1 between 1968 and 1980 (relative to 1970). An explanation may be that the shift in focus of agricultural policies towards smallholders in this period has affected commercial farmers. This theory can, however, not be confi rmed by statistics.

Agricultural characteristics
Until 2000, the agricultural sector of Zimbabwe consisted of two major farming systems, which both occupied half of the arable land (Eicher 1995 ). The commercial farming system was dominated by a relatively small group of European farmers. These large scale farms were located in the higher rainfall areas in North-Eastern Zimbabwe (Whitlow 1985 ;Matondi, 2012c ). Production was mainly focused on crops and input intensive (Whitlow 1985 ). The smallholder farming system involved a large number of African farmers. These small-scale farms were located in the drier and more remote areas with poor market access. Farming included crop and livestock production and was mainly subsistence driven (Whitlow 1985 ).
From 2000, the Zimbabwean government acquired land from the European commercial farmers on a large scale. The land was divided into smallholder farms and commercial farms of varying scales, and redistributed to black farmers (Matondi, 2012a ). Because of the limited knowledge and skills of the new farmers and poor access to inputs and new technologies, the national level of irrigation and fertilizer use declined after 2000, see Figure 2 (Matondi, 2012b ). Also, a loss of knowledge about livestock management led to more disease related deaths (Matondi, 2012b ). As a result, the number of ruminants and the ruminant density on pastures declined (Fig. A10).
At the start of the 21st century, the major farming system was maize mixed (Dixon et al. 2001 ), see the description for Zambia. As more than half of the land Figure A8 . Productiv ity developments in the Zambian agricultural and livestock sector and institutional, economic and technological/management developments. Because of limited data, agricultural tractors are not included in the inputs and in the output-input ratio for the agricultural sector.  1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 Livestock sector Index, 1961 = 100 Agricultural sector   Figure A10 . Development in livestock production intensity: ruminant density on pastures and meadows in livestock units per hectare ( LU ha −1 ), derived from FAOSTAT data ( FAOSTAT ). Ruminants included: buffaloes, camel, cattle, goats, and sheep. Note the different scale for India compared to the other countries.    In IMAGE, wheat is aggregated into the product group temperate cereals. 5 In IMAGE, soybean is aggregated into the product group oil crops. 6 In IMAGE, Australia is aggregated into the region Oceania. 7 In IMAGE, Zambia and Zimbabwe are aggregated into the region Southern Africa. 8 In MIRAGE, Zambia and Zimbabwe are aggregated into the region sub-Saharan Africa. The IMAGE model adopts yield projections from the FAO and combines these with endogenous assumptions on yield changes (Bruinsma 2003 ;OECD 2012 ). The MIRAGE model uses a baseline scenario from Aglink-Cosimo, which is also complemented by endogenous assumptions on yield developments (Laborde 2011 ). Jaggard et al. ( 2010 ) assume a continuation of current yield trends, but also take into account relative changes owing to increasing carbon dioxide (CO 2 ) and ozone (O 3 ) concentrations, climate change and technological developments.