Đang tải... (xem toàn văn)
We test an emerging theory of the forest transition using the case of Vietnam. In the early 1990s, decollectivisation of agriculture, allocation of forestry land to households, and the development of market networks transformed land use in the mountains of Vietnam, leading to an increase in forest area. We used census and geographic data covering the whole country at a fine level of aggregation to build a spatial lag regression model of reforestation. We separated natural forest regrowth from the increase in plantation forests. The forest transition theory distinguishes between the forest scarcity and economic development paths. Our study suggests that the forest scarcity path was in part at work in Vietnam: new policies allocating forestry land to households, local scarcity of forest products, and development of remote demand for timber contributed to forest cover increases. The evidence regarding the economic development path are more ambiguous, as there was no depopulation or agricultural decline in marginal regions. We propose a third forest transition path that better corresponds to Vietnam’s situation: a smallholder agricultural intensification path. In marginal regions, land scarcity associated with population growth, land degradation, and political restrictions led to, on one hand, a decline in cultivation on hillsides followed by reforestation and, on the other hand, an increase in labour inputs on the plots with the highest agro-ecological potential. The development of markets for agricultural inputs and outputs did also contribute to reforestation by raising agricultural productivity in mountain paddies and maize fields. This reinforced the concentration of agriculture on the most suitable land. r 2007 Elsevier Ltd. All rights reserved
Land Use Policy 25 (2008) 182–197 The causes of the reforestation in Vietnam Patrick Meyfroidt à , Eric F. Lambin Department of Geography, University of Louvain, Place Pasteur 3, B-1348 Louvain-La-Neuve, Belgium Received 11 December 2006; received in revised form 26 May 2007; accepted 27 June 2007 Abstract We test an emerging theory of the forest transition using the case of Vietnam. In the early 1990s, decollectivisation of agriculture, allocation of forestry land to households, and the development of market networks transformed land use in the mountains of Vietnam, leading to an increase in forest area. We used census and geographic data covering the whole country at a fine level of aggregation to build a spatial lag regression model of reforestation. We separated natural forest regrowth from the increase in plantation forests. The forest transition theory distinguishes between the forest scarcity and economic development paths. Our study suggests that the forest scarcity path was in part at work in Vietnam: new policies allocating forestry land to households, local scarcity of forest products, and development of remote demand for timber contributed to forest cover increases. The evidence regarding the economic development path are more ambiguous, as there was no depopulation or agricultural decline in marginal regions. We propose a third forest transition path that better corresponds to Vietnam’s situation: a smallholder agricultural intensification path. In marginal regions, land scarcity associated with population growth, land degradation, and political restrictions led to, on one hand, a decline in cultivation on hillsides followed by reforestation and, on the other hand, an increase in labour inputs on the plots with the highest agro-ecological potential. The development of markets for agricultural inputs and outputs did also contribute to reforestation by raising agricultural productivity in mountain paddies and maize fields. This reinforced the concentration of agriculture on the most suitable land. r 2007 Elsevier Ltd. All rights reserved. Keywords: Forest transition; Reforestation; Vietnam; Land use change; Land allocation; Forestry; Agricultural intensification; Land abandonment; Spatial lag regression Introduction Forest cover conversion and modification due to land use change impact on a range of ecosystem services (Lambin et al., 2003). While at present these processes mostly affect tropical and equatorial regions, it used to take place in Europe and other now developed parts of the world. In the past, some states succeeded in stopping deforestation and even extending their forest cover. Understanding this forest transition is attracting much interest as it provides lessons for a broader transition to sustainability (Mather, 1992). Vietnam is one of the few tropical countries where a forest trans ition seems to be taking place today. Cases of forest transition have already been studied on a local scale (Rudel et al., 2002; Klooster, 2003; Perz and Skole, 2003) and through cross-country analyses (Rudel et al., 2005; Ewers, 2006; Kauppi et al., 2006). The objective of our study was to understand the causes of refor estation in Vietnam during the 1990s on a national scale and test emerging forest transition theories (FTT) at that scale. It is based on census and geographic data at a relatively fine level of aggregation and relies on published case studies on the local scale to support interpretation. In historical analyses, Mather (1992) and Mather et al. (1999) describe a forest transition as a broad set of interrelated economic, political, institutional, and cultural processes in the agriculture, forestry, and energy sectors. Rudel et al. (2005) identify two broad pathways of forest transition: the economic development path and the forest scarcity path. The former is associated with industrialisa- tion and the tertiarisation of the economy that pull the labour force away from rural areas to cities. Agri cultural intensification reinforces this trend by increasing food ARTICLE IN PRESS www.elsevier.com/locate/landusepol 0264-8377/$ - see front matter r 2007 Elsevier Ltd. All rights reserved. doi:10.1016/j.landusepol.2007.06.001 à Corresponding author. Tel.: +32 10472675; fax: +32 10472877. E-mail addresses: patrick.meyfroidt@uclouvain.be (P. Meyfroidt), eric.lambin@uclouvain.be (E.F. Lambin). production and profitability in the most suitable regions of a country. Market networks accelerate depopulation and agricultural decline in the least suitable regions. According to this economic development path, the spatial pattern of agriculture at the na tional scale is thus expected to match increasingly land suitability. Forests are expected to regrow in marginal regions, especially those that are well connected to economic centres and to agriculturally productive regions (Mather and Needle, 1998). On a global scale, the role of agricultural intensification in this process is sometimes called the Borlaug hypothesis (Angelsen and Kaimowitz, 2001). In the second pathway of forest transition, deforestation caused by agricultural expansion or wood extraction creates a scarcity of forest products and decreases the ability of forests to deliver ecosystem goods and services. Increasing demand for wood products due to economic growth may reinforce this scarcity. The economic response by landowners includes tree planting and more intensive forest management (Hyde et al., 1996). Thanks to forestry intensification, timber needs can be satisfied from limited areas of forest plantations, thus saving the remaining forests from exploitation pressure (Sedjo and Botkin, 1997). This is similar in the forestry sector to the Borlaug hypothesis in agriculture. Forest scarcity also arouses cultural and political responses, inducing govern- ments to implement policies to restrict forest exploitation, promote mo re sustainable management practi ces, and invest in forestry research and reforestation programmes. According to this forest scarcity path, reforestation is therefore expected to occur in regions with a low forest cover, a poor land suitability for agriculture and good connections to distant markets—i.e. where developing forestry is most profitable. The substitution of fuelwood with fossil fuels or other energy sources is another possible driver of a forest transition, but with mixed empirical evidence. Recently Mather (2007) hypothesized the ex- istence of other undefined pathways of forest transition for China, India and Vietnam. Vietnam is a densely populated country, with large regional differences between small but very productive areas (the Mekong and Red River deltas) and mountai- nous areas covering more than two-thirds of the country (Fig. 1). The deltas have long been cultivated and mountains were still extens ively covered by forests until the mid-twentieth century (Poffenberger and Nguyen, 1998). Since then, Vietnam has experienced rapid defor- estation, culminating in the late 1980s. Forest cover was at its lowest in the late 1980s–early 1990s, when it covered around 25% of the territory and only 17% of the northern mountains (Figs. 2 and 3). Deforestation was mainly caused by agricultural expansion due to increasing population in the uplands, following both natural popula- tion growth and migration (De Koninck, 1999). Policies encouraged resettlement to relieve pressure on the already densely populated river deltas (Lundberg, 2004). Wood exploitation for local and urban needs also contributed to forest clearing (McElwee, 2004), as did the collectivisation of agriculture that was implemented in the north from 1954 onwards and in the south after 1975 but failed to generate sustainable agriculture (Castella and Quang, 2002). In the late 1980s, prod uctivity in mountain paddies was low because the cooperatives were poorly managed (e.g. unreliable fertiliser distribution) and farmers lacked incentives (paddy field production was the property of cooperatives). Farmers increasingly turned to slash-and- burn cultivation on hillsides to maximise labour returns while growing their own food. Agriculture is forbidden in Vietnam on most hillsides because they are classified as ‘‘forestry land’’ due to their steep slopes whether or not they have tree cover. But enforcement by local govern- ments was weak and the traditional rule of free access was prevailing on most hillsides. Food shortages became severe, forests disappeared and hill sides were increasingly eroded. Several authors warned of the deepening environmental and development crisis in Vi etnam’s mountains (Jamieson et al., 1998). Since the mid-1990s, however, this trend has been largely reversed and forest cover has increased notably through natural regeneration and the extension of tree plantations, although not everywhere in the country. This reforestation was accompanied by political and economic changes in favour of decentralisation and liberalisation (i.e. the Doi Moi reforms initiated in the 1980s). This induced rapid economic growth and the development of the industrial and service sectors (Kerkvliet and Porter, 1995). Agricultural and forestry policies changed dramatically. In the agricultural sector, input and output markets were ARTICLE IN PRESS High : 3084 Elevation (m) Low : 0 Main cities Northern mountains Central highlands 500 Kilometers Hanoi Da Nang Ho Chi Minh Cit y 0 N Fig. 1. Map of Vietnam. P. Meyfroidt, E.F. Lambin / Land Use Policy 25 (2008) 182–197 183 progressively liberalised (Pingali et al., 1997; Kerkvliet and Porter, 1995). The ‘‘Contract 100’’ policy in 1981 allowed farmers to keep surpluses above a fixed contracted quantity. The ‘‘Contract 10’’ policy in 1988 further liberalised rice and input prices, land rights, and crop choices. Households were also allowed to own all their production after subtracting taxes and charges (Kerkvliet and Porter, 1995). Under the Land Law of 1993, house- holds were given long-term rights to use, transfer, exchange, inherit, rent, and mortgage land (Do and Iyer, 2003). In mountain areas, labour was redistributed to paddies (Sadoulet et al., 2002) and the use of fertilisers and new seed varieties of rice and maize increased, with renewed incentives for farmers to produce surpluses (Minot and Goletti, 2000; Minot, 2003). Livestock and fruit trees also increased. Several local studies (Sikor, 2001; Tachiba- na et al., 2001; Castella and Er out, 2002; Muller, 2003) have pointed to the positive role of these changes on forest regeneration, but their relative contribution is still debated. In the forestry sector, tree planting campaigns were launched via Decree 327 (in 1992) and its successor, the Five Million Hectare Reforestation Programme (started in 1998) (De Jong et al., 2006). The 1993 Land Law also introduced a system for the allocation of forestry land to households (Sikor, 1998) (we use the term ‘‘forestry land’’ to refer to land legally classified for forestry use), to make the beneficiaries assume personal responsibility for protect- ing forestry land. Several authorities (forest management boards, national park administrations, stat e forest enter- prises) also signed forest protection contracts with house- holds, which imposed more restrictions on households rights on forestry land than for allocated land. Formal implementation of these policies varied locally but, in general, participating households were given rights to allocated or contracted land, such as the right to grow crops during the first years of forest regrowth and collect ARTICLE IN PRESS Fig. 2. Evolution of forest cover in Vietnam. See Meyfroidt and Lambin (to appear). 0 500 Kilometers Land cover Non forest Forest N Fig. 3. Forest cover in 1993. P. Meyfroidt, E.F. Lambin / Land Use Policy 25 (2008) 182–197184 forest products. They also received small cash payments in return for their commitment to preserve and protect forests, and sometimes to plant trees (Ministry of Agriculture and Rural Development, 2001; McElwee, 2004; Sikor, 2006; Sowerwine, 2004). The impact of these policies on forests is still under discussion. According to some studies (Tachibana et al., 2001; Castella et al., 2006), forestry land allocation led to forest recovery by prohibit- ing hillside cultivation. However, Sikor (2001, 2006) concludes that, because of the numerous shortcomings in policy design and implementation, the policy largely failed to protect uplands. Forest regeneration was thus driven by agricultural intensification. Sunderlin (2006) also presents ambiguous evidence on the impact of forestry policies. Along with agricultural and forestry policies, the Viet- namese government also developed plans to extend protected areas and strengthen their enforcement after signing the Convention on Biological Diversity (Govern- ment of Vietnam, 1994; World Conservation Monitoring Center, 1994). At the same time, the government was still promoting colonisation an d large-scale development of perennial crops (mainly coffee and rubber) in the Central Highlands (Hardy, 2000; D’haeze et al., 2005), which caused massive deforestation (De Koninck, 1999; Muller, 2003). Materials and methods Study design This study analyses the influence of agricultural and forestry changes on forests at a national scale. Actually, the forest transition theory was developed to explain change in forest cover at the country level. The units of analysis are Vietnam’s administrative districts, which are the second smallest administrative unit in Vietnam, just above the communes. A typical district is around 600 or 900 km 2 in mountainous regions. We used two regression models to identify statistical relations at the district level between forest cover changes, biophysical features, and land use variables. Local or provincial case studies were used to support interpretation of these relations. Districts were selected because they are large enough to match the concept of region as used in the FTT, while being small enough so that the findings at the district level can be interpreted using case studies at the commune (around 85 km 2 ) or village (around 10 km 2 ) levels. This multi-scale approach allows linking in-depth knowledge of processes acquired at the local level with a more statistically representative analysis at the national level. Such approach has limitations, particularly given the heterogeneous context of Vietnam’s mountains. Statistical relations do not indicate causality, and findings from case studies are only valid for the conditions from which they were derived. The study period goes from 1993 to 2002, corresponding to two forest inventories and close to two exhaustive agricultural censuses (in 1994 and 2001) that pro vide data about rural households, agricultural, and forestry land uses. This period corresponds broadly to the initial years of reforestation. Policies regarding decollectivisation and liberalisation of agriculture were largely in place in the late 1980s, but their implementation continued during the early 1990s, especially in the remote mountain regions (Minot, 2003). The consequent agricultural developments progressed throughout the 1990s. Although it was initiated earlier, forestry land allocation accelerated after the 1993 Land law. This time scale is sho rt compared to historical studies of forest transition that often extend over several decades. Therefore, some drivi ng forces identified in the FTT, such as slow demographic changes, are expected to figure less prominently in this study. However, by studying an ongoing transition, we were able to rely on more detailed data and disentangle multiple causes and pa thways of this transition. Data We assembled data from several sources in a geographi- cal information system (GIS) to perform multiple regres- sion analysis. Data were collected for the 542 districts of Vietnam. Due to missing values, we used only 344 districts in the regression analysis for natural forests, and 345 for planted forests. The following variables were included (Table 1): forest cover and forest cover change, land suitability for agriculture, accessibility to markets and population centres, agricultural intensification, agricultural and economic diversification (livestock, perennial crops such as coffee, tea and fruit trees, and non-farm sector), activity in forestry sector and its capital intensity, degree of implementation of forestry policies (which is mainly the responsibility of district authorities, Sikor, 1998), extent of protected areas, and population density. GIS layers were obtained from the Ministry of Natural Resources and Environment (MONRE), except for the Digital Elevation Model which was extracted from the Global Land One-kilometer Base Elevation (GLOBE) model from the U.S. National Oceanic and Atmospheric Administration (NOAA), at approximately 1 km resolu- tion. For forest cover, we used the forest map of 1993 produced by the Forest Inventory and Planning Institute of Vietnam (FIPI) and district-level forest statistics of 2002 from FIPI and the Forest Protection Department (FPD) (Ministry of Agriculture and Rural Development, 2003). These data were produced by supervised classification of Landsat satellite images supported by field assessment, and identify natural and planted forests. The soil map was reclassified in four suitability classes, based on Nguyen et al. (2002). We calculated the percentage of a district area covered by soils from the two lowest suitability classes. Accessibility of a district was measured by the road density and the distance to the main population centres (markets and/or administrative centres). The densities of unpa ved and paved roads were used as indicators of internal communication within a district. ARTICLE IN PRESS P. Meyfroidt, E.F. Lambin / Land Use Policy 25 (2008) 182–197 185 The main data sources on rural households and land use were the 1994 and 2001 Rural, Agricultural and Fishery Censuses conducted by the General Statistical Office of Vietnam (GSO) (General Statistical Office, 1995, 2003), covering the whole rural population. The implementation of the forestry land allocation policy was measured by the percentage of land allocated to households as forestry lands. No data on forest protection contracts were available. Forestry sector activities were proxied by the proportion of households whose main activity was in the forestry sector. The capital intensity of this sector was proxied by the number of chainsaws per person. Agricul- tural intensification was measured by the rate of increase of rice and maize yields (the two main crops), the change in the percentage of area under irrigated crops, and the change in rice cropping frequency (the number of times a plot of land is cultivated each year or, similarly, the proportion of land that is under multi-cropping). The censuses contained no district level data on rice and maize yields, livestock, and cropping frequency in paddies. For these variables, we used data at the provincial level, assuming that all districts in a province had the same value. This is likely to introduce spatial autocorrelation in our analysis, as discussed below. Data sources were Statistical Yearbooks published by the GSO (General Statistical Office, 1994, 2004), the Agricultural Census of 1994, and the General Land Census and Statistical Yearbook of 2000 (partial results were published by the Ministry of Agriculture and Rural Development, 2002). Regression models We first computed linear correlations to describe the spatial pattern of forest cover changes. We then performed ARTICLE IN PRESS Table 1 List of variables Variables used in the regression models Data sources Biophysical and accessibility variables Dtown Distance to the nearest provincial capital, in km GIS layer from MONRE Dcity Distance to the nearest large city (Ha noi, Ho Chi Minh City or Da Nang), in km GIS layer from MONRE Unpavroad Density of unpaved roads, in km=km 2 GIS layer from MONRE Pavroad Density of paved roads, in km=km 2 GIS layer from MONRE Highway Density of highways, in km=km 2 GIS layer from MONRE Slope Mean slope, in degrees NOAA-GLOBE PoorSoil % of poor quality soils GIS layer from MONRE Plateau Index of presence of plateau, calculated as mean altitude/mean slope NOAA-GLOBE Forest cover, forestry and forest policies variables For93 % of forest cover in 1993 GIS layer from FIPI Natforchg Change in % of land covered by natural forest FIPI and MARD (2003) Plaforchg Change in % of land covered by planted forest FIPI and MARD (2003) Protarea % of land assigned as protected area in 2000 GIS layer from MONRE Allocfor94 % of area allocated to households as forestry land in 1994 GSO (1995) Allocfor Change in % of area allocated to households as forestry land GSO (1995, 2003) Forhh94 % of rural households for which forestry is the main activity in 1994 GSO (1995) Forhh Change in % of rural households working primarily in forestry GSO (1995, 2003) Saw Change in number of chainsaws by 1000 rural households GSO (1995, 2003) Agricultural sector and other variables Irr Change in % of area of annual crops irrigated GSO (1995, 2003) Yldgr Growth rate of paddy rice yields (%) GSO (1994), MARD (2002) Freqgr Growth rate of paddy rice cropping frequency (%) GSO (1994), MARD (2002) MYldgr Growth rate of maize yields (%) GSO (1994), MARD (2002) Cattgr Growth rate of livestock density (%) GSO (1994), MARD (2002) Peren Change in % of area covered by perennial crops GSO (1995, 2003) Nagrhh Change in % of rural households working primarily in non-agricultural sector GSO (1995, 2003) Popden94 Rural population density in 1994 (people=km 2 ) GSO (1995, 2003) Interaction terms AllocÂslop Interaction between forestry land distributed to households and mean slope AllocÂfreq Interaction between forestry land distributed to households and growth rate of paddy cropping frequency AllocÂmyÂsl Interaction between forestry land distributed to households, growth rate of maize yields, and mean slope FreqÂslop Interaction between growth rate of rice cropping frequency and mean slope YldÂslop Interaction between growth rate of rice yields and mean slope MYldÂslop Interaction between growth rate of maize yields and mean slope PerenÂplat Interaction between change in % of area covered by perennial crops and plateau All variables are measured at the district level (unit of observation). Growth rates are annual. All change and growth rate variables are measured between 1994 and 2001, except for forest cover where the period is 1993–2002. P. Meyfroidt, E.F. Lambin / Land Use Policy 25 (2008) 182–197186 multiple linea r regressions, using a spatial lag model with a maximum likelihood estimator (MLE) (Anselin, 1988)with the Geoda software (Anselin, 2005), to account for spatial dependency in our data. We built two models where the dependent variables were specified as differences in, respectively, natural and planted forest areas between 1993 and 2002. Independent variables representing land use were also expressed as differences between 1994 and 2001, to measure how land use changes affected changes in forest cover. However, some variables were not specified as differences: biophysical and accessibility variables were assumed to be unchanged during the 10-year period of study, and initial forest cover in 1993 was included as a measure of forest abundance. In some cases, we included the variable measuring initial values along with the 2001–1994 difference: the forestry land already distributed in 1994 measured the implementation of the policy for 1993–1994, and population density and percentage of households living mainly from forestry activities in 1994 were proxies for the demand in forest products and land. The spatial lag model is expressed as y ¼ rWy þ Xb þ , with N being the number of observations and K the number of independent variables, y is a N  1 vector of the dependent varia ble, W a N  N spatial weights matrix, r the spatial lag coefficient, X a N  K matrix of independent variables, b a K  1 vector of coefficients for independent variables, and a N  1 vector of the disturbance term. This model was chosen after a diagnostic procedure whose full results are not presented here. We started with ordinary least square (OLS) regression. The presence of spatial dependency was detected by a series of Lagrange Multiplier (LM) tests (Anselin, 2005). This issue can be dealt with either by a spatial lag or a spatial error model (Anselin, 1988). Spatial error models introduce a spati al autoregressive error term. This is appropriate when error terms are spatially correlated, which is theoretically due to an unspecified variable affecting the dependent variable similarly in neighbouring locations (e.g. a provincial-level policy). LM tests rejected that model, suggesting that spatial dependency in our data was not mainly due to a misspecification. Spatial lag models are appropriate when values of the dependent variable in one location influence values of this dependent variable in neighbouring locations (e.g. due to imitation strategies of neighbouring farmers). The dependent variable in one place can be affected by surrounding values of the dependent or independent variables. A spatial lag model identifies the variables contributing to spatial dependency by verifying which parameters are affected by inclusion of the spatial autoregressive term. Parameters from spatial lag models indicate the marginal effect of a variable in one district while parameters from OLS regression indicate the aggregate effe ct of a variable, including contributions from the neighbourhood. The explanatory power in the spatial lag model can be estimated by a pseudo-R 2 , calculated as the squared correlation between observed and predicted values of the dependent variable. It is not directly comparable to the OLS R 2 . Log-Likelihood (Log-L.) values are also used to compare alternative models, including the OLS: a higher value indicates a better fit of the model. Diagnostic tests showed that the spatial lag model represented our data correctly and took the spatial dependency into accoun t. In our implementation of the spatial lag model, we used a spatial weight matrix based on distance between district centres, provided these distances were less than 67 km. Larger distances were assigned a weight of zero. This threshold was the minimum to ensure that every district had at least one neighbour. A sensitivity analysis revealed that results were robust to other thresholds (100, 150, 300, and 600 km) and that these larger thresholds did not improve the model fit based on the Log-L. criterion. We also verified that results were robust to heteroscedasticity in residuals by using HC0 standard error estimates (Long, 2000) in OLS regression. As this did not modify OLS results, we assumed it was also vali d for the MLE spatial model. Several interaction terms were computed. To analyse the differential impact of agricultural intensification and the distribution of forestry land on districts more or less suitable for agriculture, we computed the product between, on the one hand, mean slope and, on the other hand, rice cropping frequency, rate of increase of rice and maize yields, and percentage of forestry land distributed. To measure a possible synergy or redundancy between new forest policies and agricultural intensification, we com- puted the product between, on the one hand , the percentage of forestry land distributed and, on the other hand, first the cropping frequency and then the rate of increase of rice yield. For maize, we computed the product between percent age of forestry land distributed, rate of increase of yield, and mean slope, as this crop is mainly found in mountains. To test whether new forest regulations were less effective in remote areas due to weaker government enforcement capacities, we computed interac- tion terms between allocated land and, respectively, mean slope, distances to cities, and road densities. None of these latter terms were significant and they were therefore not used. Finally, we measured whether a district terrain could be characterised as a plateau by dividing mean altitude by mean slope (after adding 1 to slope to avoid division by zero). This index was multiplied by the change in area under peren nial crops to evaluate the differential impact of the development of these crops in the Central Highlands as opposed to the rest of the country. Some variables introduced notable collinearity in the model, e.g. rate of increase of rice and maize yields, and especially their interaction terms with slope and forestry policies. In each case, only one of the collinear variables was significant. It was thus retained in the regression models. Change in agricultur al mechanisation and in rural population density, and the interaction term ARTICLE IN PRESS P. Meyfroidt, E.F. Lambin / Land Use Policy 25 (2008) 182–197 187 between change in perennial crops and the presence of a plateau were also collinear. All three variables represent agricultural colonisation in the Central Highlands. We retained the interaction term and removed the two other variables. For the same reason, we did not include the accessibility variables for the model of natural forest change. Results Natural and planted forests displayed different spatial patterns of forest cover change during the study period (Figs. 4 and 5,andTables 2 and 3). Natural forest regeneration occurred mainly in the northern and central coastal mountains, in districts with high slope and thus low suitability for agriculture, and far from urban centres. Districts close to cities and in river deltas, where agriculture is most profitable, and in the Central Highlands region were less affected by reforestation or even sti ll affected by net deforestation. Tree plantations were more scattered geographically and located mainly in midlands and along coasts, where the road network is more developed, with no relation to the slope of districts. Few forest plantations were observed in peri-urban areas and in river deltas. Regression models for both natural and planted forests had a high explanatory power (Table 4). Log-L. values ARTICLE IN PRESS 0 500 Kilometer s % Change in natural forests < 0 0 0 - 10 10 - 20 > 20 N Fig. 4. Natural forest cover changes. 0 500 Kilometer s % Change in planted forests < 0 0 0 - 10 10 - 20 > 20 N Fig. 5. Planted forest cover changes. Table 2 Forest cover in 1993 and 2002 for selected regions Region Natural forests Planted forests All forests 1993 2002 1993 2002 1993 2002 Country 23.99 29.99 0.64 5.83 24.63 35.82 All mountains 29.96 40.55 0.45 5.44 30.40 45.99 Central Highlands 52.83 50.72 0.27 2.41 53.10 53.13 Northern Mountains 16.79 33.15 0.62 6.26 17.40 39.40 In % of area. ‘‘All mountains’’ refers to districts with mean slope above 2:5 . Sources: FIPI and MARD, 2003. Table 3 Pearsons’s correlations between forest cover change and geographical variables Variable Changes in natural forests Planted forests Dtown 0.15*** À0.05 Dcity 0.15*** 0.02 Unpavroad À0.11* 0.20**** Pavroad À0.13** À0.15*** Highway À0.03 0.10* Slope 0.43**** 0.05 PoorSoil 0.06 0.08* *, po0:05; **, po0:01; ***, po0:001; ***, po0:0001. P. Meyfroidt, E.F. Lambin / Land Use Policy 25 (2008) 182–197188 indicated that the spatial lag models had a better fit than the OLS models. The model for change in natural forests showed no direct influence from accessibility and biophy- sical variables at the district level. The area of remaining forest in 1993 was positively associated with forest recovery. However, increased tree planting was associated with a decline in natural forest cover. The other variables regarding forestry economy were not significant. Concern- ing forestry policies, the amount of land already allocated to households in 1994 had no explanat ory power but the area of forestry land distributed during the study period was significantly associated with natural forest regrowth. The interaction term between forest allocation and rate of increase in mountain maize yields was marginally signifi- cant and negative. Protected areas were not significant. Natural forest regeneration was not affected by change in rice cropping frequency and affected negatively by change in maize yields. However, in both cases, the interaction terms associating these variables with mean slope were highly significant and positively associated with increases in natural forests. The net effect of increases in maize yields and cropping frequency is positive for mean slopes steeper than 2:5 (for maize) and 1:4 (for rice cropping frequency). The interaction term associating the change in area covered by perennial crops with the index measuring the presence of plateau in a district terrain was negatively associated with forest cover change. The rate of increase in cattle was marginally positive and change in rural non-farm sector ARTICLE IN PRESS Table 4 Descriptive statistics and results of the spatial lag regressions for forest cover change between 1993 and 2002 Variable Descriptive statistics (units see Table 1) Spatial lag regression Mean SD Natural forest cover change Estimate Planted forest cover change Estimate Dtown 27.09 21.96 NI À0.0404* Dcity 141.92 85.84 NI À0.00127 Unpavroad 0.21 0.19 NI 6.21** Pavroad 0.17 0.26 NI À3.56 Highway 0.05 0.07 NI 22.0** Slope 2.39 2.79 À0.685 0.655 PoorSoil 14.95 17.61 0.0427 0.0192 Plateau 44.75 53.55 0.0531*** À0.0195 For93 13.21 18.50 À0.259**** À0.0699** Plaforchg 6.10 7.64 À0.189** – Protarea 3.08 8.90 0.0895 ND Forhh94 0.28 1.10 À0.651 0.538 Forhh 0.12 1.41 0.676 0.828** Saw 3.08 5.64 0.0403 À0.0518 Allocfor94 2.12 3.19 0.0401 0.608**** Allocfor 4.24 7.68 0.559**** 0.418**** AllocÂslop 19.28 38.66 NI À0.0631** AllocÂfreq 3.77 10.38 NI À0.154** Irr 2.69 10.20 À0.00539 0.00772 Yldgr 3.90 1.94 NI 0.955*** YldÂslop 11.08 14.21 NI À0.183* Freqgr 1.29 1.65 À0.649 À0.584* FreqÂslop 2.47 3.97 0.489*** 0.358** MYldgr 7.68 3.94 À0.407** NI MYldÂslop 18.98 23.74 0.164**** NI AllocÂmyÂsl 139.88 290.30 À0.00625* NI Cattgr 2.99 6.57 0.152* À0.131* Peren 2.45 3.71 À0.161 0.315* PerenÂplat 174.02 573.41 À0.00439*** À0.000549 Nagrhh À0.92 11.82 À0.00949 0.0255 Popden94 376.93 399.35 À0.00645*** À0.00546**** Spatial lag 0.307**** 0.324**** Constant 6.98** 2.03 n 344 345 Pseudo R sq. 0.58 0.52 Log-L. À1178.4 À1072.9 Log-L. of the OLS model À1185.7 À1080.8 *po0:05, **po0:01, ***po0:001, ****po0:0001. NI: variable not included in the model but non-significant (see text). ND, no determination; –; .dependent variable. P. Meyfroidt, E.F. Lambin / Land Use Policy 25 (2008) 182–197 189 was not significant. Finally, population density in 1994 was negatively related to change in natural forests and the spatial lag variable was positive and highly significant. In the model for planted forests, accessibility measures were significantly associated with positive change. Planted forests increased more rapidly in districts with higher densities of unpaved roads and highways, close to provincial capitals. Distance to the main cities was not significant. Only the density of main roads had a negative effect, perhaps as it is highest in peri-urban areas where the land rent is higher than the returns of forest plantations. Biophysical conditions were not significant. Results also showed that forests were more likely to be planted in areas with a low forest cover in 1993 and where the share of population working in forestry had increased—both a cause and a consequence. The mechanisation of forestry practices was not significant. The areas allocated to forestry already distributed to households in 1994 were related to increases in planted forests, as well as the areas of newly allocated forestry land. However, the latter effect was reduced in mountains and in districts with a higher rate of increase in rice cropping frequency. The rate of increase in paddy yields was associated with an increase in planted forests but this effect was weaker in mountain areas. The rate of increase in rice cropping frequency was negatively but very marginally associated with change in plantations. In districts with steep slopes, this association became positive and more significant. The increase in perennial crops had a positive relationship with increases in forest plantations, while an increase in cattle density had a negative one. Population density, non-agricultural popula- tion, and spatial lag parameters were related with changes in plantations in the same way as with changes in natural forests. Discussion Methodological issues Our results are robust to the model specification. However, their reliability depends on data quality. Official forest cover statistics in Vietnam have been criticised for several reasons (Lang, 2001). In a quan titative, compara- tive analysis of forest statistics and land cover maps from a variety of sources, Meyfroidt and Lambin (to appear) show that: (i) the area and spatial pattern of forest cover on the FIPI maps used in this study is consistent with the most reliable data, and (ii) these data for Vietnam indicate a turning point of forest cover around the first half of the 1990s, follo wed by an increase in forest cover of a similar magnitude as represented on the FIPI maps. Forest cover change was measured as the change in the percentage of land covered by each type of forest between 1993 and 2002. We did not use rates of forest cover change as they are sensitive to the base level (the forest cover in 1993), which was close or equal to zero for a number of districts. Changes in percentage are also more comparable with other variables, such as percentages of land allocated for forestry to households, of land classified as protected areas, etc. However, by counting onl y the net change in forest cover, the destruction of old-growth forests was aggregated with changes in secondary regrowth and therefore the quality of the new forests cannot be assessed. Similarly, it was not possible to test whether the factors decreasing deforestation were different from those promoting refor- estation (as suggested by Grainger, 1995). Finally, large negative forest cover changes can only be measured where large forest areas were present in 1993. This may partly explain why changes in natural forests were more associated (negatively) with 1993 forest cover than changes in planted forests. These problems did not arise for plante d forests, as the percentage of planted forest cover in 1993 was negligible and very few districts experienced a decrease in planted forest. In a regression analysis of the distribution of forestry land against biophysical features, accessibility, population, and land use variables (in 1994), only the slope of the terrain was significant and positive. Actually, forestry land is mainly defined by its steep slope. As we controlled for slope in the regression models of forest cover change, we assumed that the residual effect of the forestry land variable represented the policy implementation. Changes in perennial crops and in plantation forests were not correlated (Pearson’s R: 0:01, p ¼ 0:76). There was thus no overlap between these two land uses (e.g. fruit trees that may have been included in both categories). Cropping frequency was computed by combining several methods, given the high heterogeneity of farming systems in Vietnam. For the northern part of the country and the Central Highlands, with shifting cultivation and two rice cropping seasons, one being dominant, the total area sown during the year divided by the maximum area sown during one season was computed, as in Minot (2003). For the other regions with little shifting cultivation and up to three rice cropping seasons without necessarily a dominant one, the total area sown divided by the area under rice cultivation was computed, as in Minot and Goletti (2000). Results were also consis tent with those of Castella and Er out (2002). The variable for change in irrigated area accounted mainly for two rather localised changes: dykes and irrigation infrastructures were built in the southern Mekong Delta, allowing for a second crop in areas reclaimed from seasonal seawater invasion, and significant areas of irrigated agriculture were replaced by perennial crops in the eastern Red River Delta. In the rest of the country, changes in irrigated surfaces were included in the cropping frequency variable. Protected areas were initially represented by two variables: areas already protected before 1993 and newly created natural reserves between 1993 and 2000. As they displayed similar patterns, they were aggregated. The lack of explanatory power for protected areas therefore cannot be imputed to a time lag after the creation of parks. The variable on the percentage of households working primarily in the non-farm sector ARTICLE IN PRESS P. Meyfroidt, E.F. Lambin / Land Use Policy 25 (2008) 182–197190 was strongly correlated with the share of household income coming from non-agricultural sources (for 2001, Pea rson’s R: 0:74, po0:0001), which increases confidence in this indicator. The soil quality variable may be more question- able as the binary categorisation in poor versus good quality soils might be too simplistic. Finally, we lacked adequate data to measure changes in firewood consump- tion, and hydropower dams and reservoirs. Although the results suggest distinct regional dynamics (between uplands and lowlands, between northern moun- tains and Central Highlands), the analysis did not include regional dummy variables. Instead, we used continuous biophysical variables (slope an d plateau) that are more relevant to our hypotheses and that allow interpretations based on physical rather than administrative regions. Policy and economic responses to forest scarcity The main policy response to forest scarcity was the forestry land allocation to households, which restricted slash-and-burn cultivation on hillsides and provided incentives for a sound managem ent of allocated land. Sikor (2001; 2006), in a study of the actual allocation system in a northwestern commune, argued that the policy largely failed to reduce slash-and-burn cultivation because it conflicted with customary land tenure systems, thus inducing non-compliance by villagers. Local authorities were inclined to let local people, often members of their kin, cultivate their plots illegally rather than strictly enforce the rules of a physically and politically distant (due to ethnic background) government (Sikor, 2001). Investiga- tions by Nguyen (2006) and Thanh and Sikor (2006) also describe how social differentiation and the discrepancies between legal rights and actual entitlements attenuated the impact of this policy. However, despite these shortcomings, our results suggest that this policy had a positive impact on natural forest cover. Our results also suggest that forestry land allocation largely failed to stimulate the expansion of planted forests. We suspect that the association between forestry land already distributed in 1994 and the expansion of planta- tions might be spurious: districts that first started the allocation pro cedure were perhaps districts with more developed forestry activities in the first place. For an average district, the impact on plantations of forestry land distribution after 1994 is negligible, as this impact decreases in mountains where most of the allocated forestry land is situated and in areas where rice cropping frequency rose the most. This intensification occurs where growing rice is the most profitable, especially in the south- eastern region around Ho Chi Minh City. Labour was less likely to be allocated to forestry there. In mountains, the limited expansion of planted forests could be due to the absence of adequate sylvicultural knowledge and proces- sing infrastructure, and to the few market opportunities for timber compared to lowland districts. A study in hoa Binh province showed that, except for the richest farmers, tree planting was not an economically feasible option (Clement et al., 2007). In the midlands, where the infrastructure is more developed, allocated forestry land could provide profitable returns by planting trees. Other authors have also commented on the difficulty of stimulating forestry activities in remote regions because of their poor infra- structure (Lang , 2002; Ohlsson et al., 2005; Sunderlin, 2006). Actua lly, better accessibility contributed to an increase in forest plantations, which was also associated with an increase in the population working in forestry. Thus, in line with the forestry intensification argument (Hyde et al., 1996 ), demand from remote markets stimulated the growth of the forestry sector and of forest plantations in economically suitable places. However, this was not the case for natural forests. At the local level, the positive e ffect of lower forest cover on reforestation might be due to scarcity-induced increases in forest stands and the better management of natural forests by households to meet their needs, as was observed elsewhere (Hyde et al., 1996; Foster and Rosenzweig, 2003). Our results do not support the argument that the development of plantations relieves pressure on natural forests (Sedjo and Botkin, 1997). Actually, at the local level, natural forests and plantations compete for space, considering that increases in the latter were partly at the expense of the former. At an aggregated level, the reverse might be true, but our model does not allow drawing conclusions on that point. Forest regrowth was positively correlated with protected areas (Pearson’s R ¼ 0:26, po0:0001) but this association was not significant in the multiple regression. Zingerli et al. (2002) argue that, around the Ba Be park in the northern mountains, the reforesta tion observed could be attributed to agricultural changes and forestry policies rather than to park regulations that were weakly enforced. By contrast, in the study by Muller (2003) in the Central Highlands, protected areas did contribute to forest regrowth. Sowerwine et al. (1998) conclude for Ba Vi, another park in the north, that it has not been truly effective so far, but improvements in park management and better cooperation with local people might prove more successful. Cropper et al. (2001) argue that in Thailand, areas set aside often have a low suitability for agriculture and thus their status does not really increase the likelihood of their being cleared for agriculture. This may explain the positive correlation between protected areas and forest regrowth as the latter happened mostly in areas with a low suitability for agriculture. And it may also explain why protected areas are not signi ficant in the multiple regression. In conclusion, reforestation in Vietnam fits several aspects of the forest scarcity path of forest transition. New policies were implemented to address the perceived degradation of forest resources. Economic responses driven partly by urban and remote markets and probably also by a perception of local scarcity by rural households also contributed to reforestation (Mather et al., 1999; Foster ARTICLE IN PRESS P. Meyfroidt, E.F. Lambin / Land Use Policy 25 (2008) 182–197 191 [...]... on forests, depending on the region In the Central Highlands (with high values of the plateau index), the expansion of large fields of perennial crops (mainly coffee, but also tea, rubber, and ARTICLE IN PRESS P Meyfroidt, E.F Lambin / Land Use Policy 25 (2008) 182–197 other crops), spurred by market growth (urban markets in Vietnam and liberalisation of external trade), road building and government-supported... to 74 inhabitants per km2 , and from 67 to 73 in the northern mountains where reforestation was the most pronounced The number of people living off the land, even excluding forestry and fisheries, also increased (from around 5.8 to 6.3 million in the northern mountains, and from 12.1 to 13.0 million in all the mountain regions) The share of households living primarily from non-farm activities in the... not declined Agricultural intensification that has occurred has contributed to the forest transition, but mostly in the marginal regions of the country The agricultural intensification that has taken place in the most favourable regions did not drive reforestation in the marginal regions In the mountainous regions of Vietnam, our study suggests that intensification drove reforestation largely by moving labour... as driven by a growing population and diminishing returns from hillside cultivation following a ` shortening of the fallow cycle—a la Boserup Intensification was also driven by policy changes that removed barriers to increased labour inputs in paddies, and by market integration that allowed farmers to benefit from inputs There is no evidence that industrialisation or increasing trade of agricultural products...ARTICLE IN PRESS 192 P Meyfroidt, E.F Lambin / Land Use Policy 25 (2008) 182–197 and Rosenzweig, 2003) Urban and remote markets mainly stimulated the increase in plantation forests, while local scarcity stimulated the increase in both kinds of forests Similar to policies implemented in 19th century Europe and especially in France (Mather et al., 1999), the Vietnamese government restricted the use of forests... Minot, 2003) During the 1990s, maize increasingly replaced rice and cassava on hillsides, following the introduction of high-yielding varieties and the increased use of maize as fodder for livestock raised for urban markets (Sikor, 2001; Sikor and Pham, 2005; Wezel et al., 2002; Minot, 2003) In Bac Kan province, households with poor access to paddies after land allocation often adopted maize to maintain... to northwestern Vietnam This region being the poorest in the country, with few infrastructures, food security rather than forest plantations is the first priority (Minot, 2003) Together with intensification in paddy fields, the increase in maize yields was also associated with the abandonment of marginal plots in mountains In these regions, maize is the second main crop, and is used for livestock and... neighbouring districts ARTICLE IN PRESS 194 P Meyfroidt, E.F Lambin / Land Use Policy 25 (2008) 182–197 Implications for the forest transition theory While our analysis supports elements of the forest scarcity path leading to a forest transition, evidence in support of the economic development path is ambiguous The role of urbanisation and industrialisation in taking manpower out of agriculture is not measured... path of forest transition, with a redistribution of forests on marginal land, following agricultural intensification and economic development Several processes in uenced agricultural intensification in Vietnam during the 1990s The allocation of paddies to households and the liberalisation of rice trade created incentives for farmers to devote more labour per land unit (Fatoux et al., 2002), invest in land... This was reinforced by restrictions on land clearance, following new forestry policies, as shown in several case studies (Castella et al., 2006; Clement et al., 2007; Jakobsen et al., 2007) The relative profitability of slash-and-burn cultivation in the uplands therefore declined compared to the valleys The increase in cropping frequency in mountain paddies was one of the main factors contributing to natural . rate of increase in rice cropping frequency. The rate of increase in paddy yields was associated with an increase in planted forests but this effect was weaker in mountain areas. The rate of increase. Policy 25 (2008) 182–197 The causes of the reforestation in Vietnam Patrick Meyfroidt à , Eric F. Lambin Department of Geography, University of Louvain, Place Pasteur 3, B-1348 Louvain-La-Neuve,. profitability of slash-and-burn cultivation in the uplands therefore declined compared to the valleys. The increase in cropping frequency in mountain paddies was one of the main factors contributing to