Deforestation Pdf 56277 | Land Property Rights Agricultural Intensification And Deforestation In Indonesia

Partial capture of text on file.

Ecological Economics 147 (2018 ) 312–321
Contents lists available at ScienceDirect
Ecological Economics
journal homepage: www.elsevier.com/locate/ecolecon
Analysis
Land Property Rights, Agricultural Intensiﬁcation, and Deforestation in
Indonesia
Christoph Kubitzaa,⁎, Vijesh V. Krishnaa,b, Kira Urbanc, Zulkiﬂi Alamsyahd, Matin Qaima,e
aDepartment of Agricultural Economics and Rural Development, University of Goettingen, Platz der Goettinger Sieben 5, 37073 Goettingen, Germany
bInternational Maize and Wheat Improvement Center, Socioeconomics Program, Texcoco CP 56237, Mexico
cChair of Forest Inventory and Remote Sensing, University of Goettingen, Buesgenweg 5, 37077 Goettingen, Germany
dDepartment of Agricultural Economics, University of Jambi, Jalan Raya Jambi, Muara Bulian km 15, Mendalo Darat, 36361 Jambi, Indonesia
eCenter of Biodiversity and Sustainable Land Use, University of Goettingen, 37073 Goettingen, Germany
ARTICLEINFO ABSTRACT
Keywords: The expansion of agricultural land remains one of the main drivers of deforestation in tropical regions. Stronger
Rubber land property rights could possibly enable farmers to increase input intensity and productivity on the already
Land-use change cultivated land, thus reducing incentives to expand their farms by deforesting additional land. This hypothesis is
Land titles tested with data from a panel survey of farm households in Sumatra. The survey data are combined with satellite
Deforestation imageries to account for spatial patterns, such as historical forest locations. Results show that plots for which
Land-sparing agricultural intensiﬁcation farmers hold formal land titles are cultivated more intensively and are more productive than untitled plots.
Smallholders However, due to land policy restrictions, farmers located at the historic forest margins often do not hold formal
titles. Without land titles, these farmers are less able to intensify and more likely to expand into the surrounding
forest land to increase agricultural output. Indeed, forest closeness and past deforestation activities by house-
holds are found to be positively associated with current farm size. In addition to improving farmer's access to
land titles for non-forest land, better recognition of customary land rights and more eﬀective protection of forest
land without recognized claims could be useful policy responses.
1. Introduction related ecosystem functions (Steﬀan-Dewenter et al., 2007; Perfecto
and Vandermeer, 2010; Tscharntke et al., 2012). Eﬀects will vary with
Deforestation remains a widespread problem, especially in tropical the type of intensiﬁcation and also with the institutional and policy
regions. Between 2010 and 2015, about 6millionhectares of tropical context in a particular setting. Better knowledge is required about how
forest were lost annually (FAO, 2016), entailing severe negative con- land-sparing agricultural intensiﬁcation can be implemented locally,
sequences for biodiversity, ecological systems, and climate stability and why past eﬀorts have often failed. Empirical research in this di-
(Fearnside, 2005; Butler and Laurance, 2009; Wilcove et al., 2013; rection is scant.
Barnes et al., 2014). Agricultural area expansion is one of the main Here, we propose that land property rights are fundamental for
drivers of deforestation (Gibbs et al., 2010), and demand for agri- agricultural production and deforestation outcomes. Land is the main
cultural output will further increase due to population and income source of farmers' livelihoods and also a major means for accumulating
growth. In addition to food, global demand for feed, fuel, and other andinheriting wealth. The institutions shaping access, use, and transfer
biomass-derived renewable resources will grow substantially over the of land are hence central for farmers' decision-making (Deininger and
coming decades (Alexandratos and Bruinsma, 2012; Valin et al., 2014). Feder, 2001). Ownership regulations for forest land and for agricultural
Thesedevelopmentsthreatentheconservationoftheremainingtropical land often diﬀer. The available literature on the links between land
forest (Laurance et al., 2014). Increasing agricultural yields on the land property rights and deforestation focuses primarily on the eﬀects of
already cultivated, through higher input intensity and use of better secure tenure for forest land (Araujo et al., 2009; Damnyag et al., 2012;
technology, could be one important way to meet the rising demand and Liscow, 2013; Robinson et al., 2014). For agricultural land, studies have
reduce further deforestation (Green et al., 2005; Ewers et al., 2009; analyzed eﬀects of tenure security on input intensity and crop pro-
Phalan et al., 2011a; Stevenson et al., 2013). To be sure, agricultural ductivity (Deininger et al., 2011; Fenske, 2011; Bellemare, 2013), yet
intensiﬁcation is not a magic bullet to conserve tropical forest and withoutlinking this to potential deforestation outcomes. To address this
⁎Corresponding author.
E-mail address: ckubitz@gwdg.de (C. Kubitza).
https://doi.org/10.1016/j.ecolecon.2018.01.021
Received 1 June 2017; Received in revised form 10 January 2018; Accepted 17 January 2018
0921-8009/ © 2018 Elsevier B.V. All rights reserved.
C. Kubitza et al. Ecological Economics 147 (2018) 312–321
gap, we use comprehensive data from Sumatra, Indonesia, one of the claims and deforest land even when the newly obtained plots cannot be
hotspots of recent rainforest loss due to agricultural area expansion titled (Resosudarmo et al., 2014). The motivation to deforest will likely
(Margono et al., 2014; Gatto et al., 2015; Clough et al., 2016). Data increase when farmers have no titles for their already cultivated land
from a farm household survey, a village survey, and satellite imageries and therefore limited ability and incentives to intensify production.
are combined to examine relationships between land ownership rights, To answer the question whether providing secure titles for agri-
agricultural production intensity, and farm size expansion into forest cultural land could help to reduce deforestation, two sub-questions will
areas. have to be addressed. First, do land titles increase agricultural intensity
In Indonesia, small farms as well as large logging and agribusiness and productivity? Second, does higher productivity on the already
companiescontributetodeforestation (Rudel et al., 2009a; Cacho et al., cultivated land reduce farmers' incentives to clear additional forest
2014). Overall, the share of land deforested by companies is larger than land? The ﬁrst sub-question will be addressed by comparing input use
the share of land deforested by smallholder farmers. While precise data and crop productivity on farms with and without land titles and con-
are not available, smallholders may have contributed<20% to overall trolling for other relevant factors. The second sub-question is less
deforestation in Indonesia in recent decades (Lee et al., 2014). How- straightforward to answer, because this would require farm-level data
ever, there are at least two reasons why a focus on small farms – as on crop productivity in the past, which we do not have. However, we
taken in this study – is relevant nevertheless from a policy perspective. address this sub-question indirectly by analyzing the relationship be-
First, in Indonesia the role of smallholders in cultivating plantation tween the possession of land titles, historical forest coverage, defor-
crops, such as oil palm and rubber, continues to grow (Euler et al., estation activities of farm households, and farm size in a spatially ex-
2017). Second, deforestation by smallholder farms is more diﬃcult to plicit way. In addition, we look at the association between current crop
monitor and control (Krishna et al., 2017b; Kubitza et al., 2018). productivity and farm size, which – together with the other results –
Whereas large companies usually operate based on government con- may allow some cautious conclusions on the role of land titles for de-
cessions, smallholder decisions to clear forest land are individual re- forestation and the underlying mechanisms.
sponses to various incentives and constraints. Such behavioral re-
sponses need to be better understood, in order to design and implement 2. Data
eﬀective policies.
For private farms, land titles can increase agricultural intensity and 2.1. Socioeconomic Data
productivity through three eﬀects (Feder and Feeny, 1991; Besley,
1995; Deininger et al., 2011). First, the assurance eﬀect, incentivizing This research builds on data collected in Jambi Province on the is-
higher investment because farmers are more secure to also reap the land of Sumatra, Indonesia. Jambi has been one of the regions with
beneﬁts from long-term measures to improve land quality and yield rapid loss of tropical rainforest over the last few decades. Forest cover
potential. Second, the collateralization eﬀect, allowing better access to in Jambi declined from about 48% in 1990 to 30% in 2013 (Drescher
investment capital because land titles can be used as collateral in formal et al., 2016). Nevertheless, 43% of Jambi's total area was oﬃcially
credit markets. Third, the realizability eﬀect, resulting from more eﬃ- categorized as state forest in 2000 (Komarudin et al., 2008). Agri-
cient land allocation given that titled land facilitates land market cultural production in Jambi is dominated by plantation crops, espe-
transactions. The empirical literature largely conﬁrms these eﬀects cially rubber (Hevea brasiliensis) and oil palm (Elaeis guineensis). Rubber
(Banerjee et al., 2002; Goldstein and Udry, 2008; Holden et al., 2009; is primarily grown by local farmers with only some involvement of
Deininger et al., 2011; Fenske, 2011; Grimm and Klasen, 2015; Lawry large-scale companies. Companies are more involved in oil palm, but
et al., 2016), although in some cases the inﬂuence of land titling or even in oil palm>40% of the area is cultivated by smallholder farmers
more secure property rights was found to be insigniﬁcant (Quisumbing (Euler et al., 2017). That smallholders contribute to deforestation in
and Otsuka, 2001; Brasselle et al., 2002; Jacoby and Minten, 2007; Jambi in a signiﬁcant way was underlined in a recent study (Krishna
Bellemare, 2013). et al., 2017b), who showed that 18% of the rubber and oil palm plots
An increase in farm productivity induced through land titles could cultivated by smallholders were acquired through direct forest appro-
reduce deforestation (Angelsen and Kaimowitz, 2001). Higher output priation.
from the already cultivated land reduces the pressure to convert addi- AsurveyoffarmhouseholdswasconductedinJambiintworounds,
tional forest land. Also, a more productive agricultural sector could spur 2012 and 2015, as part of a larger interdisciplinary research project
broader economic development, reducing population growth, enhan- (Drescher et al., 2016). A multi-stage sampling framework was used to
cing non-agricultural income opportunities for rural households, and obtain a representative sample of local farm households. At the ﬁrst
improving land-governance capacities and institutions. Empirical evi- stage, ﬁve regencies of Jambi located in tropical lowland rainforest
dence for these types of eﬀects is scarce, although a few studies show areas were selected. At the second stage, a total of 40 villages were
indeed that higher farm productivity can help spare natural habitat randomly selected in these ﬁve regencies. In addition, ﬁve villages,
from agricultural conversion (Barbier and Burgess, 1997; Ewers et al., where more intensive measurements by other teams of the same re-
2009; Phalan et al., 2011b). On the other hand, agricultural pro- search project were ongoing (Drescher et al., 2016), were purposively
ductivity growth could also be associated with higher rates of defor- selected, resulting in a total of 45 villages. In these villages, around 700
estation, for instance, by increasing the cost of forest conservation households were randomly selected proportional to village size. There
programs or by stimulating in-migration and road infrastructure in- are two types of villages in Jambi, autochthonous and transmigrant
vestments in rural areas (Maertens et al., 2006; Phelps et al., 2013). villages. Transmigrant villages were established as part of the govern-
Better understanding the complexities in concrete situations can help ment's transmigration program (Gatto et al., 2017). Most households in
design appropriate policies aimed at promoting more sustainable de- transmigrant villages were allocated titled land by the state and started
velopment. producing plantation crops under contract with one of the large public
In Indonesia, much of the land that farmers use is not formally titled or private companies. Hence, the institutional and agricultural pro-
(Krishna et al., 2017b). Privately owned land can be titled, but the costs duction conditions are quite diﬀerent. In this research, we only consider
for farmers are relatively high. Additionally, farmers located close to the 34 autochthonous villages in the sample, with 473 farm household
the forest suﬀer from ambiguous ownership structures. Most of the observations in 2015 (and 471 household observations in 2012). Out of
forest land is formally owned by the state and not eligible for private these, around 25% are migrants (Table A1 in the Online Appendix), but
titling (Agrawal et al., 2008). But the boundaries are not always clear- these migrants in autochthonous villages did not come as part of the
cut. Some of the land that farmers have cultivated for long oﬃcially government's transmigration program (Gatto et al., 2015). Most of the
counts as forest land. Moreover, local communities have customary households in the two survey rounds are identical. The attrition rate
313
C. Kubitza et al. Ecological Economics 147 (2018) 312–321
between 2012 and 2015 was 6%. Households that could not be sur- wealth index was constructed based on ownership of the following
veyed again in 2015 (mostly due to out-migration) were replaced with assets: television, diﬀerent types of vehicles, refrigerator, and washing
other randomly selected households in the same villages. machine. A principal component analysis was used to determine the
In both survey rounds, household heads were interviewed with a weight of each asset in the wealth index (Filmer and Pritchett, 2001). μ
i
structured questionnaire, capturing a wide range of variables related to is the unobserved time-invariant heterogeneity of the model, while ε is
it
the households' socioeconomic situation and the institutional context the iid error term.
(Euler et al., 2017; Krishna et al., 2017a). Details about the diﬀerent Wealso estimate similar models at the plot level:
plots owned and cultivated by the farm households were also collected. ln(PR )=+β β LT +β X +β S +μ +ε (plot level)
pit pit it pit pit
In 2015, the 473 households cultivated a total of 902 plots with plan- 01 2 3 pi
tation crops; out of these 690 were cultivated with rubber, the rest with (2)
oil palm. For all these plots, data on general plot characteristics, such as wherePR istheannualrubberyieldperhectareonplotpofhousehold
pit
size, location, and status of land titling, were elicited. In addition, de- i at time t. LT is a dummy variable taking the value 1 if the plot was
pit
tailed input-output data were captured for all plots in 2012 and for a systematically titled at time t. S includes additional plot character-
pit
random sub-sample of plots in 2015. For the analysis of agricultural istics such as age of the rubber trees and variables related to plot lo-
productivity and intensity, we concentrate on productive rubber plots cation.
(those where the trees are old enough such that rubber is already being Due to the sampling framework used, households and plots are
harvested). Input-output relationships in rubber and oil palm are quite clustered at the village level. We account for possible heteroscedasticity
diﬀerent, so combining both crops in the same models would not make by using cluster-corrected standard errors (Pepper, 2002; Cameron
sense. Besides the interviews with household heads, village re- et al., 2011). For interpretation of the estimation coeﬃcients, func-
presentatives were interviewed in all sample villages to capture data on tional form has to be considered. SLT in Eq. (1) is a continuous vari-
it
village size, ethnic composition, and other village-level characteristics. able, so that β1 is interpreted as the percentage eﬀect on rubber yield.
LT in Eq. (2) is a dummy variable, so that the percentage eﬀects is
pit
2.2. Soil and Remote Sensing Data calculated as {exp[β −0.5×Var(β )]−1} (van Garderen and Shah,
1 1
2002).
In the farm household survey, respondents were asked to classify the The models in Eqs. (1) and (2) are estimated with random eﬀects
soil fertility on each of their plots as low, medium, or high. In addition (RE)panelestimators. Studies with micro-level data to assess the eﬀects
to these data on perceived soil quality, soil samples were taken in 2012 of land titling often struggle with endogeneity issues (Brasselle et al.,
for a randomly selected sub-sample of 92 rubber plots. These soil 2002). Endogeneity bias occurs when unobserved characteristics are
samples were taken and analyzed by a diﬀerent team of researchers jointly correlated with land titling and crop productivity. Valid in-
(Guillaume et al., 2016). We use topsoil properties, such as bulk den- struments for land titles, which are exogenous and fulﬁll the exclusion
sity, carbon content, and carbon/nitrogen ratio as additional ex- restrictions, are usually hard to ﬁnd (Fenske, 2011; Bellemare, 2013;
planatory variables in the rubber production models. Grimm and Klasen, 2015). We use diﬀerent strategies to test for en-
Land cover maps of Jambi Province from the years 1990 and 2013 dogeneity and reduce related bias to the extent possible. First, we in-
were obtained using multi-temporal Landsat TM and OLI satellite clude a wide range of plot- and household-level control variables to
imageries with a spatial resolution of 30×30m. Land cover classiﬁ- reduce the likelihood of unobserved heterogeneity. In robustness
cation is based on automatic classiﬁcation and additional qualitative, checks, we also include various measures of soil quality, which has
visual interpretation to reduce miss-classiﬁcations (Melati et al., 2014). rarely been done in previous research (Bellemare, 2013). Second, in
In this research, we are particularly interested in the share of forest in addition to using random eﬀects, we also estimate the productivity
the vicinity of the sample households, which we determined by eval- models with ﬁxed eﬀects (FE) estimators and balanced plot- and
uating land cover classiﬁcations in circles with speciﬁc radius around household-level panel data. The variation in land titling within plots
the households' residence. We use diﬀerent alternatives with 2km, and households between 2012 and 2015 is small, but suﬃcient to ob-
5km,and10kmradius.Households with a high share of forest in their tain FE estimates. We use the Hausman test (Wooldridge, 2002)to
vicinity are considered as being located at the forest margins. compare between the RE and FE models (Table A2). Test results fail to
reject the hypothesis that the RE models produce consistent estimates.
3. Econometric Methods Third, in addition to model estimates with all observations, we split the
sample into migrants and non-migrants and estimate separate models
The analysis is done in three steps. First, we present models that for these two groups. We expect heterogeneous impacts of land titling,
analyze the eﬀect of land titles on agricultural productivity. Second, we because customary land claims that apply to autochthonous people do
use similar models to analyze eﬀects of land titles on agricultural in- not apply to migrants from outside the region.
tensity (input use). Third, we examine spatial patterns by developing
and estimating models to analyze the relationships between historic 3.2. Models to Analyze Agricultural Intensity
forest margin, possession of land titles, deforestation activities, and
farm size. Toanalyzetheeﬀectoflandtitlesonintensityofrubberproduction,
we estimate plot-level panel regression models of the following type:
3.1. Models to Analyze Agricultural Productivity INV =+β β LT +β X +β S +μ +ε (plot level)
pit pit it pit pit
01 2 3 pi
To analyze the eﬀect of land titles on productivity in rubber, we (3)
estimate household-level panel regression models of the following type: ln(LS) =+β β LT +β X +β S +μ +ε (plot level)
pit pit it pit pit
ln(PR ) =+β β SLT +β X +μ +ε (household level) 01 2 3 pi
it it it it (1)
01 2 i (4)
where PR is total annual rubber yield per hectare of household i at whereINV istotalannualexpenditures on material inputs applied per
it pit
time t. SLT is the share of household i's land cultivated with plantation hectare (ha) on plot p by household i at time t. Material inputs include
it
crops that had a systematic land title at time t. The share can vary chemical fertilizers and pesticides (incl. herbicides). LSpit is annual
between 0 and 1. Xit is a vector of other farm and household char- labor input (incl. family and hired labor) measured in hours per ha. The
acteristics that may also inﬂuence rubber yields, such as farm size, age, other variables are deﬁned as above. Since>50% of the sample
gender, and education of the household head, and a wealth index. The farmers did not use any material inputs during the survey years, we do
314
C. Kubitza et al. Ecological Economics 147 (2018) 312–321
Fig. 1. Maps of land uses in Jambi Province (Sumatra) in 1990 and 2013.
Notes: Maps 1 and 2 depict Jambi Province in 1990 and 2013. Map 3 is one example from a sub-region (Harapan Rainforest) with eight sample villages in 1990. The red circles indicate a
2kmradius around the sample households' residence. Circles with diﬀerent radius (2, 5, 10km) were used to calculate the share of forest land around households.
not take logs of INV and use a linear functional form instead. Given hold land titles and therefore have stronger incentives to expand their
pit
censoring of the dependent variable at 0, we use a Tobit speciﬁcation farms into the forest. After controlling for other factors, this should lead
for the model in Eq. (3). To test the eﬀect of INV and LS on crop to larger farm sizes at the forest margins. To test this hypothesis, we
pit pit
productivity, we also estimate additional speciﬁcations of Eq. (2) with regress farm size in 2015 on forest closeness in 1990 and a set of control
these inputs included as explanatory variables. variables. Again, we used Moran's I, Anselin's, and Florax's Lagrange
Multiplier tests (Baltagi, 2003) to test for spatial autocorrelation. These
3.3. Spatial Regression Models tests reject the hypothesis of zero spatial autocorrelation, so we esti-
mate spatial lag models of the following type:
To estimate the eﬀect of historical forest closeness on the prob- ln(FS ) =+ρWln(FS ) β +β F +β V +β V
iv iv iv iv v
ability of holding a land title, we estimate the following plot-level 01 2 3
probit model: +ε (household level) (6)
P(LT ) =+β β F +β Z +β Z +β Z +ε (plot level) iv
piv iv piv iv v piv where FS is total farm size of household i in village v measured in
01 2 3 4 iv
(5) hectares, F is the share of forest land in 1990 (as deﬁned above). V
iv iv
and V are household- and village-level controls. W is an N×N spatial
where LT is a dummy indicating whether or not plot p of household i v
piv weights matrix (N=number of households) based on the inverse Eu-
in village v was systematically titled in 2015, and Fiv is the share of clidian distance between the households' residence. The parameter ρ
forest land in 1990 in a circle with speciﬁc radius around the household measures the degree of spatial correlation. W is row standardized, such
residence. F can take values between 0 (no forest in 1990) to 1
iv that for each i,∑ w = 1(Baltagi, 2003). The spatial lag ρWln(FS ) can
(completely forested in 1990). The reference year 1990 was chosen j ij iv
because most of the formal land classiﬁcations in Indonesia took place be interpreted as a weighted average of the farm sizes of neighboring
in the 1980s (Indrarto et al., 2012). We estimate separate models, using households. For comparison, spatial error and ordinary least squares
radii of 2km, 5km, and 10km to construct F . In each of these models, models are reported in Table A4.
iv
plots that are located outside the speciﬁc radius are excluded from es-
timation. A further robustness check is performed, replacing F with a 4. Results
iv
binary variable indicating if the plot was acquired by the household
through deforestation. Z , Z , and Z are further plot-, household-, and 4.1. Descriptive Statistics
piv iv v
village-level controls. Eq. (5) includes both rubber and oil palm plots.
It is likely that land titling is also aﬀected by spatial factors such as The average size of farms in our sample in 2015 was around 4ha.
local policies, distances to roads and markets, or environmental con- This refers to the land cultivated, regardless of whether or not the
ditions. This can possibly lead to spatial dependency in the models in farmer formally owns the diﬀerent plots. Locations of the farm house-
Eq. (5). All models were tested for spatial autocorrelation using Moran's holds are depicted in Fig. 1 (Maps 1 and 2). Responses during the
I, Anselin's, and Florax's Lagrange Multiplier tests (Baltagi, 2003). These survey interviews suggest that households are actively engaged in de-
tests failed to reject the hypothesis of zero spatial autocorrelation. For forestation. This is also conﬁrmed by land cover maps. In 1990, about
completeness, spatial lag and spatial error models are reported in Table 17% of the area within a 5km radius around farmers' residence was
A3. coveredwithforest; by2013, this forest share was reduced to 3%. Much
Wehypothesize that households close to the forest are less likely to of the previous forest land is now grown with rubber and oil palm. Even
315

The words contained in this file might help you see if this file matches what you are looking for:

...Ecological economics contents lists available at sciencedirect journal homepage www elsevier com locate ecolecon analysis land property rights agricultural intensication and deforestation in indonesia christoph kubitzaa vijesh v krishnaa b kira urbanc zulkii alamsyahd matin qaima e adepartment of rural development university goettingen platz der goettinger sieben germany binternational maize wheat improvement center socioeconomics program texcoco cp mexico cchair forest inventory remote sensing buesgenweg ddepartment jambi jalan raya muara bulian km mendalo darat ecenter biodiversity sustainable use articleinfo abstract keywords the expansion remains one main drivers tropical regions stronger rubber could possibly enable farmers to increase input intensity productivity on already change cultivated thus reducing incentives expand their farms by deforesting additional this hypothesis is titles tested with data from a panel survey farm households sumatra are combined satellite imageries a...

Related files

Share

Help

Related files

Share

Share to social media

Help

Login Area