- Email: [email protected]
Combining national forest type maps with annual global tree cover maps to better understand forest change over time: Case study for Thailand
Applied Geography 62 (2015) 294e300
Contents lists available at ScienceDirect
Applied Geography journal homepage: www.elsevier.com/locate/apgeog
Combining national forest type maps with annual global tree cover maps to better understand forest change over time: Case study for Thailand Brian A. Johnson Institute for Global Environmental Strategies, 2108-11, Kamiyamaguchi, Hayama, Kanagawa, 240-0115, Japan
a r t i c l e i n f o
a b s t r a c t
Article history: Received 3 April 2015 Received in revised form 26 May 2015 Accepted 26 May 2015 Available online xxx
National and global land use/land cover (LULC)/LULC change (LULCC) data sets often have different strengths and weaknesses for monitoring forest change over time. For example, a national-level map may be very detailed in terms of number and type of forest-related LULC classes, but infrequently updated compared to a global map with fewer LULC classes (e.g. percent tree cover maps or forest/nonforest maps). So, additional useful information might be gained by integrating national and global LULC data sets. As a demonstration, in this study a national forest type map of Thailand from the year 2000 was combined with annual global tree cover maps for the years 2000e2012 to obtain multi-temporal information on forest change in Thailand and to create a baseline estimate of forest change to 2020 (i.e. with no new policy interventions). Results showed that all forest types experienced declines in area from 2000 to 2012, with the greatest area losses for Mixed Deciduous Forests (137,765 ha) and the greatest percentage losses for Swamp Forests (5.8%). Annual forest losses, in general, increased at a near-linear rate from 2000 to 2012, and are projected to increase from 39,290 ha/year in 2012 to 51,775 ha/year by the end of 2015 (an increase of 31.8%) and 66,945 ha/year by 2020 (an increase of 70.4%) based on linear extrapolation of the historical trend. For comparison, net forest loss is currently around 5,211,000 ha/year at the global level and 677,000 ha/year at the South and Southeast Asia regional level (Food and Agriculture Organization of the United Nations, 2010b). The methods presented here provide a computationally-simple approach to annually update existing forest maps and estimate future forest change using free global tree cover data. © 2015 Elsevier Ltd. All rights reserved.
Keywords: Forest monitoring Data fusion Deforestation Land use change Landsat REDDþ
1. Introduction 1.1. National and global land use/land cover change mapping efforts related to forest monitoring The loss and degradation of natural forests has significant impacts on the broader terrestrial, atmospheric, and aquatic environments, and Southeast Asia has been identified as a region with high natural forest loss (Sodhi, Koh, Brook, & Ng, 2004; Sodhi et al., 2010). Frequent measurements of natural forest extent are needed to assist national/sub-national land use planning, monitor compliance with international environmental agreements (e.g. Kyoto Protocol, Convention on Biological Diversity), and implement conservation-based incentive programs (e.g. REDDþ). To support
E-mail address: [email protected]. http://dx.doi.org/10.1016/j.apgeog.2015.05.011 0143-6228/© 2015 Elsevier Ltd. All rights reserved.
these and other efforts, many national and global land use/land cover change (LULCC) mapping studies have been conducted using satellite imagery and remote sensing image processing techniques. Previous LULCC studies vary greatly in terms of their mapping methodologies (e.g. automated vs. manual mapping approaches, choice of classification algorithm for automated mapping), spatial resolutions, temporal resolutions, classification systems (i.e. number/type of LULC classes), and classification accuracies, and many examples of different approaches can be seen in the Country Reports of FAO's (2010a, b) Global Forest Resources Assessment (FRA) (http://www.fao.org/forestry/fra/67090/en/, last accessed May 11, 2015). This variety of methodologies makes it difficult to aggregate the national-level maps for regional-to-global scale assessments (Food and Agriculture Organization of the United Nations, 2010b; Hansen, Stehman, & Potapov, 2010). So, a major advantage of global LULCC maps are their high consistency across space (Hansen et al., 2013). On the other hand, national-level LULCC mapping
B.A. Johnson / Applied Geography 62 (2015) 294e300
studies have typically been conducted using finer spatial resolution imagery (e.g. 30 m or finer in many of the FAO Country Reports) than global mapping studies, which have mainly used imagery with spatial resolutions of 250 m or coarser (Bellot, Bertram, Navratil, Siegert, & Dotzauer, 2014; Bontemps et al., 2011; Friedl et al., 2010; Hansen et al., 2003; Hansen et al., 2010; Wang et al., 2014), so often the national maps could detect finer-scale changes that the global maps could not. The use of the finer resolution data for global LULC mapping (and LULCC mapping) is difficult due to the limited availability of high-quality (e.g. cloud-free) imagery covering the entire land surface, the high spectral and textual variability of global landscapes, and the high computation requirements (Chen et al., 2015). Only in the last couple of years have global LULCC studies been conducted at resolutions of 30 m (Chen et al., 2015; Hansen et al., 2013) or finer (Shimada et al., 2014), and these fine resolution LULCC products still have some limitations compared to many national LULCC data sets, such as lower temporal resolution (e.g. only two years of LULC maps for Chen et al. (2015)) or less detailed land use information (e.g. only one (Hansen et al., 2013) or two (Shimada et al., 2014) LULC classes). Given the different strengths and weaknesses of national and global LULCC mapping efforts, there seem to be some possibilities to combine national and global data sets for more effective LULCC monitoring and/or modeling, especially if the data sets have similar spatial resolutions. For example, if a national data set has very detailed LULC information (e.g. many LULC classes) but a low temporal resolution, it may be useful to combine it with a global data set having a higher temporal resolution (but less detailed LULC information) to better assess the rates of change of at least some types of LULC of interest (e.g. forests). In comparison, previous studies on LULC map integration have mainly focused on combining existing maps to generate a more accurate single-date LULC map (Clinton, Yu, & Gong, 2015; Schepaschenko et al., 2011; Song et al., 2014). To demonstrate how existing national and global LULC data sets can be combined for improved forest change monitoring, here a national LULC map of Thailand (Wichawutipong, 2006) having very detailed forest type information (12 forest type classes) but a low temporal resolution (only one map date) was combined with a global tree cover data set (Hansen et al., 2013) having less detailed LULC information (percent tree cover/tree cover change) but a high temporal resolution (annual maps). By combining the two data sets, multi-temporal information on deforestation/forest degradation (DFD) was obtained, and a baseline level of future LULCC related to forests was generated to 2020. To the author's knowledge, this is the first study to combine existing national and global LULC products to predict future LULCC. 1.2. Overview of annual global tree cover (AGTC) maps and Thailand forest type map The annual global tree cover (AGTC) maps recently produced by Hansen et al. (2013) have a great potential for monitoring forest cover and forest cover change at national, regional, and global scales due to their high spatial (30 m) and temporal resolutions (annual maps from 2000 to 2012 using a temporally-consistent methodology). However, these maps do not differentiate between natural forest and other tree habitats (e.g. plantations), or between different natural forest types, which have different implications for climate change (Intergovernmental Panel on Climate Change, 2006), biodiversity (Schnittler & Stephenson, 2000), and other ecosystem services (Sohngen & Brown, 2006). For example, Tropical Evergreen Forests and Mixed Deciduous Forests provide habitat for different species (Forest Carbon Partnership Facility, 2013; Roy & Tomar, 2000) and have different levels of carbon storage
295
(Forest Carbon Partnership Facility, 2013; Intergovernmental Panel on Climate Change, 2006), so for habitat modeling or quantification of ecosystem services, it is necessary to understand the forest area/ forest area change by forest type. Additionally, the AGTC maps have been found to incorrectly show non-forest and/or non-tree lands (e.g. shrublands and agricultural lands containing pineapple, soybeans, tea, bananas, etc.) as containing tree cover/tree losses/tree gains due to automated mapping errors (Bellot et al., 2014; Tropek, 2014), leading to significant overestimates of forest extent and forest loss (i.e. “phantom deforestation”) (Bellot et al., 2014). To overcome the limitations of the AGTC maps, a previous study combined the AGTC maps with a global “Intact Forest Landscapes” map (Potapov et al., 2008) (i.e. a map of forested areas 50,000 ha or larger with no significant human activity) and a landform map of Indonesia (i.e. a map of wetland, other lowland, upland, and montane formations) to assess primary forest loss by landform type in Indonesia (Margono, Potapov, Turubanova, Stolle, & Hansen, 2014). However, the previous work did not take into account smaller forested areas (only 52 of the total 38,672 natural forest areas in Thailand were 50,000 ha) and did not discriminate between the different forest types which may exist within a single landform type. In many cases it is useful to monitor changes in all natural forested areas rather than just large primary forests, as small, fragmented forest patches can also provide important ecosystem services (e.g. habitat provisioning for animals that € , Norman, pollinate crops and disperse seeds) (Bodin, Tengo Lundberg, & Elmqvist, 2006). As discussed in Section 1.1., many countries prepare detailed national land use and/or forest maps, and these data sets can potentially be integrated with the AGTC maps (or other global maps) for improved LULCC monitoring. As one example, Thailand's Royal Forestry Department produced a detailed national forest type map for the year 2000 in which polygons of 12 different forest types were manually-digitized based on Landsat TM image interpretation and ground verification (Wichawutipong, 2006). The polygons in this map delineate the general outer boundaries of the forest lands, but some areas within these boundaries lack tree cover due to human (e.g. logging) or natural (e.g. fire) disturbances, and these non-tree areas do not provide the same ecosystem services as the tree-covered areas, e.g. they provide less erosion protection and runoff mitigation (United States Department of Agriculture, 1986) and store less carbon (Intergovernmental Panel on Climate Change, 2006), so the ecosystem services provided by forests would be overestimated if these non-tree areas are not removed. The AGTC maps could be used to separate these tree and non-tree areas within the areas designated as natural forest. The national forest type map has not been updated since the year 2000, but more general forest/non-forest maps of Thailand (i.e. maps without forest type information) have been produced (Food and Agriculture Organization of the United Nations, 2010a; The Royal Forestry Department of Thailand, 2013). However, the use of different types of imagery (with different spatial resolutions) and/or different definitions of forest in different years makes it difficult to assess trends in forest change over time (Forest Carbon Partnership Facility, 2013). The lack of forest type information in the forest maps produced since 2000 also reduces their usefulness for many types of environmental analysis (e.g. habitat modeling, climate modeling, ecosystem service quantification). To summarize, the main limitations of the AGTC maps as related to this study is their lack of detailed LULC information and the main limitation of the national forest type map is its lack of update since the year 2000. These two data sets have the same spatial resolution (30 m, both being produced from Landsat imagery), but different numbers of LULC classes and temporal resolutions, so they seem to contain complementary information for LULCC monitoring. For
296
B.A. Johnson / Applied Geography 62 (2015) 294e300
example, since LULCC maps from multiple points in time are typically needed for modeling future LULCC (Huettner, Leemans, Kok, & Ebeling, 2009), the AGTC data set can be used to provide multitemporal information to the national forest type map and allow for future forest change prediction. Based on this logic, in this study the two data sets were combined to assess historical trends in natural forest change. Unlike previous work (Margono et al., 2014), here small forested areas were also considered, and a baseline estimate of future forest change (to 2015 and 2020) was derived based on the historical trend. Baseline forest change estimates can be used to calculate avoided forest losses due to policy interventions or incentive programs, and are required for some payment for ecosystem services programs (e.g. REDDþ) (United Nations Framework Convention on Climate Change, 2009). The methods presented here can be replicated quite easily in other countries that have forest type maps with Landsat-like spatial resolution, and they provide a simple approach for assessing historical forest change and predicting future changes. 2. Methods
increasing trend in annual NFTC loss, while a positive slope indicates a decreasing trend), while the short-term trends were assessed using a 5-year moving average (line pointing down indicates NFTC loss is increasing in more recent years, while line pointing up indicates NFTC loss is decreasing). Forest types with a high risk of future DFD were defined as those with an increasing trend in annual NFTC loss over both the long- and short-terms, forest types with a medium risk of future DFD were defined as those with an increasing long-term trend but a flat/decreasing short-term trend in NFTC loss (because DFD lessened in recent years), and forest types with a low risk of future DFD were defined as those with a flat/decreasing trend in NFTC loss over both the long- and short-terms. Linear extrapolation was used to generate a baseline level of future NFTC change (for all natural forest lands, in general) to 2020, based on the annual trend from 2000 to 2012. Linear extrapolation is a relatively common method for predicting future LULCC based on historical LULC maps (typically using LULC maps from at least three points in time), and it is also referred to as the “Spatial Historical Approach” (SpHA) for LULCC modeling by Huettner et al. (2009).
2.1. Combining the national and global maps 3. Results The Thailand forest type map for the year 2000 was downloaded from http://www.savgis.org/thailand.htm#THAILANDUSE (last accessed 01 April 2015), and the AGTC maps (version 1.0) for the 2000e2012 period were downloaded from http:// earthenginepartners.appspot.com/science-2013-global-forest/download_v1.0.html (last accessed 01 April 2015). The Thailand forest type map is actually a land use map which also contains information on non-forest land use types (e.g. agriculture, plantation, urban, etc.), but the non-forest land use types were excluded from this study because they did not contain natural forest cover according to the Royal Forestry Department's mapping methodology. The AGTC data consisted of three raster maps generated using automated image processing techniques (Hansen et al., 2013): (i) a “tree canopy cover” map for the year 2000 in which pixel values represent the canopy closure of vegetation taller than 5 m in height, (ii) a “year of gross forest cover loss” map in which pixel values represent the year of disturbance (i.e. full tree cover loss within the pixel), and (iii) a binary “forest cover gain” map in which pixel values represent gain or no gain in tree cover. To combine the national and global data sets, they were projected to the same geographic coordinate system (World Geodetic System 1984) and overlaid onto one another, and the pixels from the AGTC maps located within “natural forest” land use polygons were extracted using Geographic Information Systems (GIS) software. After combining the data sets, natural forest tree cover (NFTC) for the year 2000 (NFTC_2000) was defined as the pixels in map (i) containing tree cover and located within “natural forest” land use polygons. NFTC losses and gains were defined as the loss or gain of tree cover, respectively, within “natural forest” land use polygons. Annual net losses and gains in NFTC were calculated assuming equal annual gains in tree cover because the “forest cover gain map” did not specify the year of tree cover gains. Finally, NFTC_2000 and NFTC losses and gains were converted from pixel units to area in ha. 2.2. Assessing deforestation/forest degradation risks and predicting future forest change Annual trends in NFTC change by forest type were analyzed to assess the future DFD risks of each forest type. The longer-term trend in NFTC changes were assessed over the entire 12-year period (negative slope of a simple linear regression indicates an
3.1. Visual comparison of the maps before and after combination As can be seen in Fig. 1 (b), some areas designated as “natural forest” land use in the national map (Dry Dipterocarp forest in the Fig.) did not actually contain tree cover due to the reasons given in Section 1.2. Also shown in Fig. 1 (b), several non-forest areas in the AGTC maps contained tree cover (e.g. Eucalyptus plantations and agricultural lands). Since the aim in this study was to monitor NFTC change over time, the non-tree covered areas within natural forests and the tree-covered areas outside natural forests needed to be excluded. Combining the two data sets (Fig. 1 (c)) allowed for these areas to be successfully excluded, as only the tree-covered areas within the natural forest land use polygons were extracted. 3.2. NFTC change from 2000 to 2012 As shown in Table 1, the total NFTC of Thailand decreased from 15,921,297 ha in 2000 to 15,609,451 ha in 2012; a net change of 311,847 ha. (1.96%). The average net NFTC change over the period was 25,987 ha/year, significantly less than FAO's estimate of forest change over a similar time period: 63,000 ha/year from 2000 to 2010 (Food and Agriculture Organization of the United Nations, 2010a). FAO's higher estimate of forest loss is likely due to the inclusion of plantations used for wood-based products in their forest cover/forest change calculations (Food and Agriculture Organization of the United Nations, 2010a). Differences may also have arisen because FAO's forest change estimates for 2000e2010 were based on linear extrapolation using forest maps from two years (2000 and 2004); fewer than the recommended number for this type of analysis (three or more) (Huettner et al., 2009). 3.3. NFTC change from 2000 to 2012 by forest type Net changes in NFTC varied significantly by forest type, as shown in Table 1, as did the annual trends in net NFTC change, shown in Fig. 3. Some forest types had a high loss of NFTC over the 2000e2012 period due to an extreme natural event in just one year. For example, Beach Forest had a high relative NFTC loss of 3.75% over the 2000e2012 period, but the annual NFTC changes show that nearly all of the losses occurred in 2005, after the 2004 Indian Ocean tsunami destroyed many coastal areas (net gains in tree
B.A. Johnson / Applied Geography 62 (2015) 294e300
297
Fig. 1. Dry Dipterocarp forest and Eucalyptus plantation in the year 2000 (from the national land use map) (a), tree cover map for the year 2000 (from the annual global tree cover maps), shown in light green (b), and extracted natural forest tree cover after integrating the national land use map and the annual global tree cover maps (c). Landsat ETM þ satellite image from 07 December, 1999 shown as basemap. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.).
(77,286 ha), and Tropical Evergreen (48,214 ha), while the three forest types with the greatest percentage tree cover losses were Swamp (5.79%), Dry Dipterocarp (4.44%), and Beach Forest (3.75%). The forest types with the highest risk of future DFD were Mixed Deciduous, Dry Dipterocarp, and Inundated. Since Mixed Deciduous and Dry Dipterocarp were also two of the most dominant forest types in Thailand, the risk of future DFD to all the country's natural forested land, in general, is also categorized as high in Fig. 3. 3.4. Predicted NFTC change from 2013 to 2020
Fig. 2. Annual net NFTC change (including all forest types) from 2001 to 2012. Actual changes shown for 2001e2012 as well as predicted changes for 2013e2020 (indicated by the solid line). “Correlation” indicates the R value of a simple linear regression using the 2001e2012 data.
cover occurred in most other years). So, in this case, the net NFTC change over the period gives a misleading impression of the future DFD risk, because other large tsunamis are probably unlikely to occur again in the near future. This example demonstrates why it is important to consider LULC at three or more points in time for the prediction of future LULCC (as a third LULC map post-2005 would have shown NFTC gain for Beach Forest since 2005). The three forest types with the greatest tree cover losses by area were Mixed Deciduous (137,765 ha), Dry Dipterocarp
The trend in tree cover loss for all natural forests, in general, was near-linear (R ¼ 0.868) over the time period, as shown in Fig. 2. Based on linear extrapolation, the predicted total NFTC net losses reach 51,775 ha/year by the end of 2015 (an increase of 31.8%) and 66,945 ha/year by the end of 2020 (an increase of 70.4%) if no additional policy intervention occurs. Based on these estimates, the total NFTC of Thailand is projected to decrease from 15.921 Mha (year 2000) to 15.545 Mha by the end of 2015 and to 15.401 Mha by the end of 2020. Although more sophisticated methods than linear extrapolation exist for projecting future natural forest change, e.g. spatially-explicit LULCC models (Echeverria, Coomes, Hall, & Newton, 2008; Huettner et al., 2009), their higher complexity tends to limit their transparency and clarity to policy-makers (Huettner et al., 2009), and the greater expertise/higher computation effort required to run these models makes them difficult to apply at the national level in many countries (Huettner et al., 2009).
Table 1 Change in natural forest tree cover (NFTC) from 2000 to 2012, by forest type.*“NF land use” includes tree and non-tree areas.
NF land use 2000 (Mha.)* NFTC 2000 (Mha.) NFTC 2012 (Mha.) Gross loss (ha.) Gross gain (ha.) Net change (ha.) Net change (%) Average net change per year (%)
Mixed Dry Dry Tropical Hill Secondary Bamboo Mangrove Pine deciduous evergreen dipterocarp evergreen evergreen
Swamp Inundated Beach
All natural forest
8.613 8.450 8.312 162,908 25,143 137,765 1.63% 0.14%
0.029 0.027 0.026 1699 118 1581 5.79% 0.48%
16.297 15.921 15.609 375,747 63,900 311,847 1.96% 0.16%
2.240 2.210 2.186 29,912 6322 23,591 1.07% 0.09%
1.830 1739 1.662 91,184 13,899 77,286 4.44% 0.37%
1.461 1.405 1.357 56,227 8013 48,214 3.43% 0.29%
1.423 1406 1.394 17,794 5716 12,079 0.86% 0.07%
0.278 0.270 0.266 7354 2504 4851 1.79% 0.15%
0.147 0.145 0.144 2018 682 1336 0.92% 0.08%
0.119 0.188 0.185 4522 902 3620 1.92% 0.16%
0.046 0.046 0.045 484 198 287 0.63% 0.05%
0.024 0.233 0.225 1053 195 858 3.67% 0.31%
0.011 0.010 0.098 592 212 380 3.75% 0.31%
298
B.A. Johnson / Applied Geography 62 (2015) 294e300
Fig. 3. Annual net NFTC change (in ha.; shown on y-axis) by forest type. Black line shows the 5-year moving average. High, moderate, and low indicate the risk of future forest losses based on long-term (12 year) and short-term (5 year) trends in forest change.
Here, linear extrapolation was used for illustration simplicity and to maintain consistency with the approach used in Thailand's FAO Country Report (2010a), but other LULCC models could be used instead, if desired. Alternatively, an adjustment to the linearlyextrapolated forest change estimate could be made taking into account national circumstances (e.g. based on a national development plan), as suggested in Thailand's UN-REDD readiness preparation proposal (Forest Carbon Partnership Facility, 2013). 4. Discussion 4.1. Possible counter-measures to reduce future forest loss The main deforestation drivers in Thailand have been previously identified as the conversion of natural forests to agriculture (food and energy crops), infrastructure development, and mining, while the main forest degradation drivers are illegal commercial logging, illegal commercial harvesting of non-timber products, and forest fires (Forest Carbon Partnership Facility, 2013). The previous report by the Forest Carbon Partnership Facility (2013) found that these drivers did not vary significantly by agro-ecological region, but that claim seems to be contradicted by the findings of this study, as forest change rates varied widely by forest type (which vary by agro-ecological region). Because of the varying change rates of different forest types, climate change-related forest conservation policies and/or incentive programs might be made more effective by increasing their focus on the forest types with high future DFD risk and wide area coverage (e.g. Mixed Deciduous and Dry Dipterocarp) to prevent future losses in NFTC from reaching the projected levels. This focus may be particularly important due to relatively low conservation priority currently given to Dry Dipterocarp Forests in Southeast Asia in comparison to other forest types like Tropical Evergreen and Mangrove (Wohlfart, Wegmann, & Leimgruber, 2014) (both of
which showed only moderate future DFD risks in Thailand in this study). In previous research, payment for ecosystem services (PES) schemes that target areas at greater risk of deforestation have been found more efficient than more general (non-targeted) forestrelated PES schemes (Alix-Garcia, De Janvry, & Sadoulet, 2008), so this may be one option for conserving forests with high DFD risk on privately-owned lands. Biodiversity-related policies/incentive programs might instead focus policies/incentives on the forest types with high DFD risk and high relative tree cover losses (e.g. Swamp or Inundated) to ensure that rarer forest types (which may also contain rarer plant and animal species) are given proper consideration. 4.2. Uncertainties The main sources of uncertainty in this study are: (1) potential overestimation or underestimation of tree cover extent, losses, and gains within “natural forest” land use polygons, due to the automated mapping errors in the AGTC maps, (2) temporal errors in the tree cover loss year (e.g. if cloud cover prevented the detection of a tree cover loss until the following year), which would affect the analysis of annual trends in tree cover change, (3) potential overestimation of NFTC gains if non-natural tree cover (e.g. Oil palm or Eucalyptus plantations) is established on previously cleared areas in the natural forest land use polygons, and (4) potential underestimation of NFTC gains if natural forest gains occur outside the natural forest land use polygons. Uncertainties related to (1) and (2) are inherent to the AGTC data set (and not the proposed method), so they can only be overcome by improving the AGTC automated mapping methodology. Uncertainties (3) and (4) are due to the low temporal resolution of the national land use map, so to minimize these uncertainties and allow for better future forest change estimation, the national land use map would need to be updated periodically (e.g. every 10e15 years). For example, NFTC gains detected
B.A. Johnson / Applied Geography 62 (2015) 294e300
within “natural forest” land use polygons could be removed if an updated forest type map showed that the tree cover gains were due to Oil palm planting rather than natural forest regeneration. On the other hand, if an updated forest type map shows a newlyestablished natural forest, NFTC gains within this new forested area could be included for the historical trend analysis. An additional benefit of updating the land use map would be that the processes driving LULCC in each area of forest loss could be identified (e.g. conversion of forest to agriculture, urban expansion). 5. Conclusions A single-date national forest type map from the year 2000 was integrated with annual global tree cover (AGTC) maps to monitor the annual changes in Thailand's natural forest cover from 2000 to 2012. Thailand's natural forest losses were found to have increased at a near-linear rate over the period, reaching 39,290 ha/year by 2012, and natural forest losses are predicted to increase to 51,775 ha/year by the end of 2015 (an increase of 31.8%) and 66,945 ha/year by 2020 (an increase of 70.4%) if no additional countermeasures are taken. Based on analysis of the long- and short-term trends in annual forest loss by forest type, the forest types with the highest risks of future deforestation/forest degradation (DFD) were identified as Mixed Deciduous, Dry Dipterocarp, and Inundated. Since Mixed Deciduous and Dry Dipterocarp are two of the most dominant forest types in Thailand, reducing the DFD of these forest types is critical to prevent natural forest losses from reaching the predicted levels. The AGTC maps used in this study are planned to be updated annually in the future, and another annual global mapping effort (“forest/non-forest” maps) with similar spatial resolution (25 m) has been started by the Japan Aerospace Exploration Agency (Shimada et al., 2014) (presently limited to the 2007e2010 period, but also planned for updating). These annual tree cover/forest cover maps, when integrated with more detailed (but less frequently updated) national- or sub-national level forest type data sets, have great potential for assessing DFD trends to enable better decisionmaking on how to achieve forest conservation goals related to climate change, biodiversity, and ecosystem services, and the methods presented in this study can serve as a template for other countries' implementation of spatially- and temporally-consistent forest monitoring. Finally, in cases where national data sets with more detailed and/or more accurate forest type information are not available, high resolution (~30 m) global land use/land cover maps containing more general forest information, e.g. the GlobeLand30 maps for the years 2000 and 2010 (Chen et al., 2015), can alternatively be integrated with AGTC maps to assess annual trends in forest change. Acknowledgments This paper is generally based upon outputs produced through projects on “monitoring, reporting, and verification” and “new market mechanisms”, funded by the Japanese Ministry of Environment. References Alix-Garcia, J., De Janvry, A., & Sadoulet, E. (2008). The role of deforestation risk and calibrated compensation in designing payments for environmental services. Environment and Development Economics, 13(3), 375e394. Bellot, F., Bertram, M., Navratil, P., Siegert, F., & Dotzauer, H. (2014). The highresolution global map of 21st-century forest cover change from the University of Maryland (“Hansen Map”) is hugely overestimating deforestation in Indonesia. Retrieved from http://www.forclime.org/documents/press_release/ FORCLIME_Overestimation of Deforestation.pdf. €, M., Norman, A., Lundberg, J., & Elmqvist, T. (2006). The value of Bodin, O., Tengo
299
small size: loss of forest patches and ecological thresholds in southern Madagascar. Ecological Applications, 16(2), 440e451. Bontemps, S., Defourny, P., Bogaert, E., Van, Arino, O., Kalogirou, V., & Perez, J. R. (2011). Globcover 2009: Products description and validation report (pp. 1e53). Retrieved from http://due.esrin.esa.int/globcover/LandCover2009/GLOBCOVER2009_Validation_ Report_2.2.pdf. Chen, J., Chen, J., Liao, A., Cao, X., Chen, L., Chen, X., et al. (2015). Global land cover mapping at 30m resolution: a POK-based operational approach. ISPRS Journal of Photogrammetry and Remote Sensing, 103, 7e27. http://dx.doi.org/10.1016/ j.isprsjprs.2014.09.002. Clinton, N., Yu, L., & Gong, P. (2015). Geographic stacking: decision fusion to increase global land cover map accuracy. ISPRS Journal of Photogrammetry and Remote Sensing, 103, 57e65. http://dx.doi.org/10.1016/j.isprsjprs.2015.02.010. Echeverria, C., Coomes, D. A, Hall, M., & Newton, A. C. (2008). Spatially explicit models to analyze forest loss and fragmentation between 1976 and 2020 in southern Chile. Ecological Modelling, 212, 439e449. Food and Agriculture Organization of the United Nations. (2010a). Global forest resources assessment 2010 country report: Thailand. Food and Agriculture Organization of the United Nations. (2010b). Global forest resources assessment 2010: Main report. Forest Carbon Partnership Facility. (2013). Readiness preparation proposal (R-PP) for Thailand. Retrieved from http://www.forestcarbonpartnership.org/sites/fcp/ files/2013/RRevised RePP version 24 Feb 2013.pdf. Friedl, M. A., Sulla-Menashe, D., Tan, B., Schneider, A., Ramankutty, N., Sibley, A., et al. (2010). MODIS collection 5 global land cover: algorithm refinements and characterization of new datasets. Remote Sensing of Environment, 114(1), 168e182. http://dx.doi.org/10.1016/j.rse.2009.08.016. Hansen, M. C., DeFries, R. S., Townshend, J. R. G., Carroll, M., Dimiceli, C., & Sohlberg, R. A. (2003). Global percent tree cover at a spatial resolution of 500 meters: first results of the MODIS vegetation continuous fields algorithm. Earth Interactions, 7, 1e15. Hansen, M. C., Potapov, P. V., Moore, R., Hancher, M., Turubanova, S. A, Tyukavina, A., et al. (2013). High-resolution global maps of 21st-century forest cover change. Science, 342, 850e853. Hansen, M. C., Stehman, S. V., & Potapov, P. V. (2010). Quantification of global gross forest cover loss. Proceedings of the National Academy of Sciences of the United States of America, 107(19), 8650e8655. Huettner, M., Leemans, R., Kok, K., & Ebeling, J. (2009). A comparison of baseline methodologies for “Reducing emissions from deforestation and degradation.” Carbon Balance and Management, 4, 4. Intergovernmental Panel on Climate Change. (2006). Agriculture, forestry, and other land use. In 2006 IPCC guidelines for national greenhouse gas inventories (Vol. 4) http://dx.doi.org/10.1016/j.phrs.2011.03.002. Margono, B. A., Potapov, P. V., Turubanova, S., Stolle, F., & Hansen, M. C. (2014). Primary forest cover loss in Indonesia over 2000e2012. Nature Climate Change, 4, 730e735. Potapov, P., Yaroshenko, A., Turubanova, S., Dubinin, M., Laestadius, L., Thies, C., et al. (2008). Mapping the world's intact forest landscapes by remote sensing. Ecology and Society, 13(2), 51. Roy, P. S., & Tomar, S. (2000). Biodiversity characterization at landscape level using geospatial modelling technique. Biological Conservation, 95(1), 95e109. Schepaschenko, D., McCallum, I., Shvidenko, A., Fritz, S., Kraxner, F., & Obersteiner, M. (2011). A new hybrid land cover dataset for Russia: a methodology for integrating statistics, remote sensing and in situ information. Journal of Land Use Science, 6(4), 245e259. http://dx.doi.org/10.1080/ 1747423X.2010.511681. Schnittler, M., & Stephenson, S. L. (2000). Myxomycete biodiversity in four different forest types in Costa Rica. Mycologia, 92(4), 626e637. Shimada, M., Itoh, T., Motooka, T., Watanabe, M., Shiraishi, T., Thapa, R., et al. (2014). New global forest/non-forest maps from ALOS PALSAR data (2007e2010). Remote Sensing of Environment, 155, 13e31. Sodhi, N. S., Koh, L. P., Brook, B. W., & Ng, P. K. L. (2004). Southeast Asian biodiversity: an impending disaster. Trends in Ecology and Evolution, 19(12), 654e660. Sodhi, N. S., Posa, M. R. C., Lee, T. M., Bickford, D., Koh, L. P., & Brook, B. W. (2010). The state and conservation of Southeast Asian biodiversity. Biodiversity and Conservation, 19, 317e328. Sohngen, B., & Brown, S. (2006). The influence of conversion of forest types on carbon sequestration and other ecosystem services in the South Central United States. Ecological Economics, 57, 698e708. Song, X.-P., Huang, C., Feng, M., Sexton, J. O., Channan, S., & Townshend, J. R. (2014). Integrating global land cover products for improved forest cover characterization: an application in North America. International Journal of Digital Earth, 7(9), 709e724. http://dx.doi.org/10.1080/17538947.2013.856959. The Royal Forestry Department of Thailand. (2013). Forestry statistics data 2013. Tropek, R. (2014). Comment on “High-resolution global maps of 21st-century forest cover change. Science, 344, 981. http://dx.doi.org/10.1126/ science.1248753. United Nations Framework Convention on Climate Change. (2009). Decision 4/CP.15. United States Department of Agriculture. (1986). Urban hydrology for small watersheds. In Technical release (Vol. 55). Wang, J., Zhao, Y., Li, C., Yu, L., Liu, D., & Gong, P. (2014). Mapping global land cover in 2001 and 2010 with spatial-temporal consistency at 250m resolution. ISPRS Journal of Photogrammetry and Remote Sensing, 103, 38e47. http://dx.doi.org/ 10.1016/j.isprsjprs.2014.03.007.
300
B.A. Johnson / Applied Geography 62 (2015) 294e300
Wichawutipong, J. (2006). Forest area monitoring in Thailand with the use of satellite imagery. In 49th session of the committee on the peaceful uses of outer space. Vienna, Austria. Retrieved from http://www.unoosa.org/pdf/pres/copuos2006/ symposium/06.pdf.
Wohlfart, C., Wegmann, M., & Leimgruber, P. (2014). Mapping threatened dry deciduous dipterocarp forest in South-east Asia for conservation management. Tropical Conservation Science, 7(4), 597e613.
Contents lists available at ScienceDirect
Applied Geography journal homepage: www.elsevier.com/locate/apgeog
Combining national forest type maps with annual global tree cover maps to better understand forest change over time: Case study for Thailand Brian A. Johnson Institute for Global Environmental Strategies, 2108-11, Kamiyamaguchi, Hayama, Kanagawa, 240-0115, Japan
a r t i c l e i n f o
a b s t r a c t
Article history: Received 3 April 2015 Received in revised form 26 May 2015 Accepted 26 May 2015 Available online xxx
National and global land use/land cover (LULC)/LULC change (LULCC) data sets often have different strengths and weaknesses for monitoring forest change over time. For example, a national-level map may be very detailed in terms of number and type of forest-related LULC classes, but infrequently updated compared to a global map with fewer LULC classes (e.g. percent tree cover maps or forest/nonforest maps). So, additional useful information might be gained by integrating national and global LULC data sets. As a demonstration, in this study a national forest type map of Thailand from the year 2000 was combined with annual global tree cover maps for the years 2000e2012 to obtain multi-temporal information on forest change in Thailand and to create a baseline estimate of forest change to 2020 (i.e. with no new policy interventions). Results showed that all forest types experienced declines in area from 2000 to 2012, with the greatest area losses for Mixed Deciduous Forests (137,765 ha) and the greatest percentage losses for Swamp Forests (5.8%). Annual forest losses, in general, increased at a near-linear rate from 2000 to 2012, and are projected to increase from 39,290 ha/year in 2012 to 51,775 ha/year by the end of 2015 (an increase of 31.8%) and 66,945 ha/year by 2020 (an increase of 70.4%) based on linear extrapolation of the historical trend. For comparison, net forest loss is currently around 5,211,000 ha/year at the global level and 677,000 ha/year at the South and Southeast Asia regional level (Food and Agriculture Organization of the United Nations, 2010b). The methods presented here provide a computationally-simple approach to annually update existing forest maps and estimate future forest change using free global tree cover data. © 2015 Elsevier Ltd. All rights reserved.
Keywords: Forest monitoring Data fusion Deforestation Land use change Landsat REDDþ
1. Introduction 1.1. National and global land use/land cover change mapping efforts related to forest monitoring The loss and degradation of natural forests has significant impacts on the broader terrestrial, atmospheric, and aquatic environments, and Southeast Asia has been identified as a region with high natural forest loss (Sodhi, Koh, Brook, & Ng, 2004; Sodhi et al., 2010). Frequent measurements of natural forest extent are needed to assist national/sub-national land use planning, monitor compliance with international environmental agreements (e.g. Kyoto Protocol, Convention on Biological Diversity), and implement conservation-based incentive programs (e.g. REDDþ). To support
E-mail address: [email protected]. http://dx.doi.org/10.1016/j.apgeog.2015.05.011 0143-6228/© 2015 Elsevier Ltd. All rights reserved.
these and other efforts, many national and global land use/land cover change (LULCC) mapping studies have been conducted using satellite imagery and remote sensing image processing techniques. Previous LULCC studies vary greatly in terms of their mapping methodologies (e.g. automated vs. manual mapping approaches, choice of classification algorithm for automated mapping), spatial resolutions, temporal resolutions, classification systems (i.e. number/type of LULC classes), and classification accuracies, and many examples of different approaches can be seen in the Country Reports of FAO's (2010a, b) Global Forest Resources Assessment (FRA) (http://www.fao.org/forestry/fra/67090/en/, last accessed May 11, 2015). This variety of methodologies makes it difficult to aggregate the national-level maps for regional-to-global scale assessments (Food and Agriculture Organization of the United Nations, 2010b; Hansen, Stehman, & Potapov, 2010). So, a major advantage of global LULCC maps are their high consistency across space (Hansen et al., 2013). On the other hand, national-level LULCC mapping
B.A. Johnson / Applied Geography 62 (2015) 294e300
studies have typically been conducted using finer spatial resolution imagery (e.g. 30 m or finer in many of the FAO Country Reports) than global mapping studies, which have mainly used imagery with spatial resolutions of 250 m or coarser (Bellot, Bertram, Navratil, Siegert, & Dotzauer, 2014; Bontemps et al., 2011; Friedl et al., 2010; Hansen et al., 2003; Hansen et al., 2010; Wang et al., 2014), so often the national maps could detect finer-scale changes that the global maps could not. The use of the finer resolution data for global LULC mapping (and LULCC mapping) is difficult due to the limited availability of high-quality (e.g. cloud-free) imagery covering the entire land surface, the high spectral and textual variability of global landscapes, and the high computation requirements (Chen et al., 2015). Only in the last couple of years have global LULCC studies been conducted at resolutions of 30 m (Chen et al., 2015; Hansen et al., 2013) or finer (Shimada et al., 2014), and these fine resolution LULCC products still have some limitations compared to many national LULCC data sets, such as lower temporal resolution (e.g. only two years of LULC maps for Chen et al. (2015)) or less detailed land use information (e.g. only one (Hansen et al., 2013) or two (Shimada et al., 2014) LULC classes). Given the different strengths and weaknesses of national and global LULCC mapping efforts, there seem to be some possibilities to combine national and global data sets for more effective LULCC monitoring and/or modeling, especially if the data sets have similar spatial resolutions. For example, if a national data set has very detailed LULC information (e.g. many LULC classes) but a low temporal resolution, it may be useful to combine it with a global data set having a higher temporal resolution (but less detailed LULC information) to better assess the rates of change of at least some types of LULC of interest (e.g. forests). In comparison, previous studies on LULC map integration have mainly focused on combining existing maps to generate a more accurate single-date LULC map (Clinton, Yu, & Gong, 2015; Schepaschenko et al., 2011; Song et al., 2014). To demonstrate how existing national and global LULC data sets can be combined for improved forest change monitoring, here a national LULC map of Thailand (Wichawutipong, 2006) having very detailed forest type information (12 forest type classes) but a low temporal resolution (only one map date) was combined with a global tree cover data set (Hansen et al., 2013) having less detailed LULC information (percent tree cover/tree cover change) but a high temporal resolution (annual maps). By combining the two data sets, multi-temporal information on deforestation/forest degradation (DFD) was obtained, and a baseline level of future LULCC related to forests was generated to 2020. To the author's knowledge, this is the first study to combine existing national and global LULC products to predict future LULCC. 1.2. Overview of annual global tree cover (AGTC) maps and Thailand forest type map The annual global tree cover (AGTC) maps recently produced by Hansen et al. (2013) have a great potential for monitoring forest cover and forest cover change at national, regional, and global scales due to their high spatial (30 m) and temporal resolutions (annual maps from 2000 to 2012 using a temporally-consistent methodology). However, these maps do not differentiate between natural forest and other tree habitats (e.g. plantations), or between different natural forest types, which have different implications for climate change (Intergovernmental Panel on Climate Change, 2006), biodiversity (Schnittler & Stephenson, 2000), and other ecosystem services (Sohngen & Brown, 2006). For example, Tropical Evergreen Forests and Mixed Deciduous Forests provide habitat for different species (Forest Carbon Partnership Facility, 2013; Roy & Tomar, 2000) and have different levels of carbon storage
295
(Forest Carbon Partnership Facility, 2013; Intergovernmental Panel on Climate Change, 2006), so for habitat modeling or quantification of ecosystem services, it is necessary to understand the forest area/ forest area change by forest type. Additionally, the AGTC maps have been found to incorrectly show non-forest and/or non-tree lands (e.g. shrublands and agricultural lands containing pineapple, soybeans, tea, bananas, etc.) as containing tree cover/tree losses/tree gains due to automated mapping errors (Bellot et al., 2014; Tropek, 2014), leading to significant overestimates of forest extent and forest loss (i.e. “phantom deforestation”) (Bellot et al., 2014). To overcome the limitations of the AGTC maps, a previous study combined the AGTC maps with a global “Intact Forest Landscapes” map (Potapov et al., 2008) (i.e. a map of forested areas 50,000 ha or larger with no significant human activity) and a landform map of Indonesia (i.e. a map of wetland, other lowland, upland, and montane formations) to assess primary forest loss by landform type in Indonesia (Margono, Potapov, Turubanova, Stolle, & Hansen, 2014). However, the previous work did not take into account smaller forested areas (only 52 of the total 38,672 natural forest areas in Thailand were 50,000 ha) and did not discriminate between the different forest types which may exist within a single landform type. In many cases it is useful to monitor changes in all natural forested areas rather than just large primary forests, as small, fragmented forest patches can also provide important ecosystem services (e.g. habitat provisioning for animals that € , Norman, pollinate crops and disperse seeds) (Bodin, Tengo Lundberg, & Elmqvist, 2006). As discussed in Section 1.1., many countries prepare detailed national land use and/or forest maps, and these data sets can potentially be integrated with the AGTC maps (or other global maps) for improved LULCC monitoring. As one example, Thailand's Royal Forestry Department produced a detailed national forest type map for the year 2000 in which polygons of 12 different forest types were manually-digitized based on Landsat TM image interpretation and ground verification (Wichawutipong, 2006). The polygons in this map delineate the general outer boundaries of the forest lands, but some areas within these boundaries lack tree cover due to human (e.g. logging) or natural (e.g. fire) disturbances, and these non-tree areas do not provide the same ecosystem services as the tree-covered areas, e.g. they provide less erosion protection and runoff mitigation (United States Department of Agriculture, 1986) and store less carbon (Intergovernmental Panel on Climate Change, 2006), so the ecosystem services provided by forests would be overestimated if these non-tree areas are not removed. The AGTC maps could be used to separate these tree and non-tree areas within the areas designated as natural forest. The national forest type map has not been updated since the year 2000, but more general forest/non-forest maps of Thailand (i.e. maps without forest type information) have been produced (Food and Agriculture Organization of the United Nations, 2010a; The Royal Forestry Department of Thailand, 2013). However, the use of different types of imagery (with different spatial resolutions) and/or different definitions of forest in different years makes it difficult to assess trends in forest change over time (Forest Carbon Partnership Facility, 2013). The lack of forest type information in the forest maps produced since 2000 also reduces their usefulness for many types of environmental analysis (e.g. habitat modeling, climate modeling, ecosystem service quantification). To summarize, the main limitations of the AGTC maps as related to this study is their lack of detailed LULC information and the main limitation of the national forest type map is its lack of update since the year 2000. These two data sets have the same spatial resolution (30 m, both being produced from Landsat imagery), but different numbers of LULC classes and temporal resolutions, so they seem to contain complementary information for LULCC monitoring. For
296
B.A. Johnson / Applied Geography 62 (2015) 294e300
example, since LULCC maps from multiple points in time are typically needed for modeling future LULCC (Huettner, Leemans, Kok, & Ebeling, 2009), the AGTC data set can be used to provide multitemporal information to the national forest type map and allow for future forest change prediction. Based on this logic, in this study the two data sets were combined to assess historical trends in natural forest change. Unlike previous work (Margono et al., 2014), here small forested areas were also considered, and a baseline estimate of future forest change (to 2015 and 2020) was derived based on the historical trend. Baseline forest change estimates can be used to calculate avoided forest losses due to policy interventions or incentive programs, and are required for some payment for ecosystem services programs (e.g. REDDþ) (United Nations Framework Convention on Climate Change, 2009). The methods presented here can be replicated quite easily in other countries that have forest type maps with Landsat-like spatial resolution, and they provide a simple approach for assessing historical forest change and predicting future changes. 2. Methods
increasing trend in annual NFTC loss, while a positive slope indicates a decreasing trend), while the short-term trends were assessed using a 5-year moving average (line pointing down indicates NFTC loss is increasing in more recent years, while line pointing up indicates NFTC loss is decreasing). Forest types with a high risk of future DFD were defined as those with an increasing trend in annual NFTC loss over both the long- and short-terms, forest types with a medium risk of future DFD were defined as those with an increasing long-term trend but a flat/decreasing short-term trend in NFTC loss (because DFD lessened in recent years), and forest types with a low risk of future DFD were defined as those with a flat/decreasing trend in NFTC loss over both the long- and short-terms. Linear extrapolation was used to generate a baseline level of future NFTC change (for all natural forest lands, in general) to 2020, based on the annual trend from 2000 to 2012. Linear extrapolation is a relatively common method for predicting future LULCC based on historical LULC maps (typically using LULC maps from at least three points in time), and it is also referred to as the “Spatial Historical Approach” (SpHA) for LULCC modeling by Huettner et al. (2009).
2.1. Combining the national and global maps 3. Results The Thailand forest type map for the year 2000 was downloaded from http://www.savgis.org/thailand.htm#THAILANDUSE (last accessed 01 April 2015), and the AGTC maps (version 1.0) for the 2000e2012 period were downloaded from http:// earthenginepartners.appspot.com/science-2013-global-forest/download_v1.0.html (last accessed 01 April 2015). The Thailand forest type map is actually a land use map which also contains information on non-forest land use types (e.g. agriculture, plantation, urban, etc.), but the non-forest land use types were excluded from this study because they did not contain natural forest cover according to the Royal Forestry Department's mapping methodology. The AGTC data consisted of three raster maps generated using automated image processing techniques (Hansen et al., 2013): (i) a “tree canopy cover” map for the year 2000 in which pixel values represent the canopy closure of vegetation taller than 5 m in height, (ii) a “year of gross forest cover loss” map in which pixel values represent the year of disturbance (i.e. full tree cover loss within the pixel), and (iii) a binary “forest cover gain” map in which pixel values represent gain or no gain in tree cover. To combine the national and global data sets, they were projected to the same geographic coordinate system (World Geodetic System 1984) and overlaid onto one another, and the pixels from the AGTC maps located within “natural forest” land use polygons were extracted using Geographic Information Systems (GIS) software. After combining the data sets, natural forest tree cover (NFTC) for the year 2000 (NFTC_2000) was defined as the pixels in map (i) containing tree cover and located within “natural forest” land use polygons. NFTC losses and gains were defined as the loss or gain of tree cover, respectively, within “natural forest” land use polygons. Annual net losses and gains in NFTC were calculated assuming equal annual gains in tree cover because the “forest cover gain map” did not specify the year of tree cover gains. Finally, NFTC_2000 and NFTC losses and gains were converted from pixel units to area in ha. 2.2. Assessing deforestation/forest degradation risks and predicting future forest change Annual trends in NFTC change by forest type were analyzed to assess the future DFD risks of each forest type. The longer-term trend in NFTC changes were assessed over the entire 12-year period (negative slope of a simple linear regression indicates an
3.1. Visual comparison of the maps before and after combination As can be seen in Fig. 1 (b), some areas designated as “natural forest” land use in the national map (Dry Dipterocarp forest in the Fig.) did not actually contain tree cover due to the reasons given in Section 1.2. Also shown in Fig. 1 (b), several non-forest areas in the AGTC maps contained tree cover (e.g. Eucalyptus plantations and agricultural lands). Since the aim in this study was to monitor NFTC change over time, the non-tree covered areas within natural forests and the tree-covered areas outside natural forests needed to be excluded. Combining the two data sets (Fig. 1 (c)) allowed for these areas to be successfully excluded, as only the tree-covered areas within the natural forest land use polygons were extracted. 3.2. NFTC change from 2000 to 2012 As shown in Table 1, the total NFTC of Thailand decreased from 15,921,297 ha in 2000 to 15,609,451 ha in 2012; a net change of 311,847 ha. (1.96%). The average net NFTC change over the period was 25,987 ha/year, significantly less than FAO's estimate of forest change over a similar time period: 63,000 ha/year from 2000 to 2010 (Food and Agriculture Organization of the United Nations, 2010a). FAO's higher estimate of forest loss is likely due to the inclusion of plantations used for wood-based products in their forest cover/forest change calculations (Food and Agriculture Organization of the United Nations, 2010a). Differences may also have arisen because FAO's forest change estimates for 2000e2010 were based on linear extrapolation using forest maps from two years (2000 and 2004); fewer than the recommended number for this type of analysis (three or more) (Huettner et al., 2009). 3.3. NFTC change from 2000 to 2012 by forest type Net changes in NFTC varied significantly by forest type, as shown in Table 1, as did the annual trends in net NFTC change, shown in Fig. 3. Some forest types had a high loss of NFTC over the 2000e2012 period due to an extreme natural event in just one year. For example, Beach Forest had a high relative NFTC loss of 3.75% over the 2000e2012 period, but the annual NFTC changes show that nearly all of the losses occurred in 2005, after the 2004 Indian Ocean tsunami destroyed many coastal areas (net gains in tree
B.A. Johnson / Applied Geography 62 (2015) 294e300
297
Fig. 1. Dry Dipterocarp forest and Eucalyptus plantation in the year 2000 (from the national land use map) (a), tree cover map for the year 2000 (from the annual global tree cover maps), shown in light green (b), and extracted natural forest tree cover after integrating the national land use map and the annual global tree cover maps (c). Landsat ETM þ satellite image from 07 December, 1999 shown as basemap. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.).
(77,286 ha), and Tropical Evergreen (48,214 ha), while the three forest types with the greatest percentage tree cover losses were Swamp (5.79%), Dry Dipterocarp (4.44%), and Beach Forest (3.75%). The forest types with the highest risk of future DFD were Mixed Deciduous, Dry Dipterocarp, and Inundated. Since Mixed Deciduous and Dry Dipterocarp were also two of the most dominant forest types in Thailand, the risk of future DFD to all the country's natural forested land, in general, is also categorized as high in Fig. 3. 3.4. Predicted NFTC change from 2013 to 2020
Fig. 2. Annual net NFTC change (including all forest types) from 2001 to 2012. Actual changes shown for 2001e2012 as well as predicted changes for 2013e2020 (indicated by the solid line). “Correlation” indicates the R value of a simple linear regression using the 2001e2012 data.
cover occurred in most other years). So, in this case, the net NFTC change over the period gives a misleading impression of the future DFD risk, because other large tsunamis are probably unlikely to occur again in the near future. This example demonstrates why it is important to consider LULC at three or more points in time for the prediction of future LULCC (as a third LULC map post-2005 would have shown NFTC gain for Beach Forest since 2005). The three forest types with the greatest tree cover losses by area were Mixed Deciduous (137,765 ha), Dry Dipterocarp
The trend in tree cover loss for all natural forests, in general, was near-linear (R ¼ 0.868) over the time period, as shown in Fig. 2. Based on linear extrapolation, the predicted total NFTC net losses reach 51,775 ha/year by the end of 2015 (an increase of 31.8%) and 66,945 ha/year by the end of 2020 (an increase of 70.4%) if no additional policy intervention occurs. Based on these estimates, the total NFTC of Thailand is projected to decrease from 15.921 Mha (year 2000) to 15.545 Mha by the end of 2015 and to 15.401 Mha by the end of 2020. Although more sophisticated methods than linear extrapolation exist for projecting future natural forest change, e.g. spatially-explicit LULCC models (Echeverria, Coomes, Hall, & Newton, 2008; Huettner et al., 2009), their higher complexity tends to limit their transparency and clarity to policy-makers (Huettner et al., 2009), and the greater expertise/higher computation effort required to run these models makes them difficult to apply at the national level in many countries (Huettner et al., 2009).
Table 1 Change in natural forest tree cover (NFTC) from 2000 to 2012, by forest type.*“NF land use” includes tree and non-tree areas.
NF land use 2000 (Mha.)* NFTC 2000 (Mha.) NFTC 2012 (Mha.) Gross loss (ha.) Gross gain (ha.) Net change (ha.) Net change (%) Average net change per year (%)
Mixed Dry Dry Tropical Hill Secondary Bamboo Mangrove Pine deciduous evergreen dipterocarp evergreen evergreen
Swamp Inundated Beach
All natural forest
8.613 8.450 8.312 162,908 25,143 137,765 1.63% 0.14%
0.029 0.027 0.026 1699 118 1581 5.79% 0.48%
16.297 15.921 15.609 375,747 63,900 311,847 1.96% 0.16%
2.240 2.210 2.186 29,912 6322 23,591 1.07% 0.09%
1.830 1739 1.662 91,184 13,899 77,286 4.44% 0.37%
1.461 1.405 1.357 56,227 8013 48,214 3.43% 0.29%
1.423 1406 1.394 17,794 5716 12,079 0.86% 0.07%
0.278 0.270 0.266 7354 2504 4851 1.79% 0.15%
0.147 0.145 0.144 2018 682 1336 0.92% 0.08%
0.119 0.188 0.185 4522 902 3620 1.92% 0.16%
0.046 0.046 0.045 484 198 287 0.63% 0.05%
0.024 0.233 0.225 1053 195 858 3.67% 0.31%
0.011 0.010 0.098 592 212 380 3.75% 0.31%
298
B.A. Johnson / Applied Geography 62 (2015) 294e300
Fig. 3. Annual net NFTC change (in ha.; shown on y-axis) by forest type. Black line shows the 5-year moving average. High, moderate, and low indicate the risk of future forest losses based on long-term (12 year) and short-term (5 year) trends in forest change.
Here, linear extrapolation was used for illustration simplicity and to maintain consistency with the approach used in Thailand's FAO Country Report (2010a), but other LULCC models could be used instead, if desired. Alternatively, an adjustment to the linearlyextrapolated forest change estimate could be made taking into account national circumstances (e.g. based on a national development plan), as suggested in Thailand's UN-REDD readiness preparation proposal (Forest Carbon Partnership Facility, 2013). 4. Discussion 4.1. Possible counter-measures to reduce future forest loss The main deforestation drivers in Thailand have been previously identified as the conversion of natural forests to agriculture (food and energy crops), infrastructure development, and mining, while the main forest degradation drivers are illegal commercial logging, illegal commercial harvesting of non-timber products, and forest fires (Forest Carbon Partnership Facility, 2013). The previous report by the Forest Carbon Partnership Facility (2013) found that these drivers did not vary significantly by agro-ecological region, but that claim seems to be contradicted by the findings of this study, as forest change rates varied widely by forest type (which vary by agro-ecological region). Because of the varying change rates of different forest types, climate change-related forest conservation policies and/or incentive programs might be made more effective by increasing their focus on the forest types with high future DFD risk and wide area coverage (e.g. Mixed Deciduous and Dry Dipterocarp) to prevent future losses in NFTC from reaching the projected levels. This focus may be particularly important due to relatively low conservation priority currently given to Dry Dipterocarp Forests in Southeast Asia in comparison to other forest types like Tropical Evergreen and Mangrove (Wohlfart, Wegmann, & Leimgruber, 2014) (both of
which showed only moderate future DFD risks in Thailand in this study). In previous research, payment for ecosystem services (PES) schemes that target areas at greater risk of deforestation have been found more efficient than more general (non-targeted) forestrelated PES schemes (Alix-Garcia, De Janvry, & Sadoulet, 2008), so this may be one option for conserving forests with high DFD risk on privately-owned lands. Biodiversity-related policies/incentive programs might instead focus policies/incentives on the forest types with high DFD risk and high relative tree cover losses (e.g. Swamp or Inundated) to ensure that rarer forest types (which may also contain rarer plant and animal species) are given proper consideration. 4.2. Uncertainties The main sources of uncertainty in this study are: (1) potential overestimation or underestimation of tree cover extent, losses, and gains within “natural forest” land use polygons, due to the automated mapping errors in the AGTC maps, (2) temporal errors in the tree cover loss year (e.g. if cloud cover prevented the detection of a tree cover loss until the following year), which would affect the analysis of annual trends in tree cover change, (3) potential overestimation of NFTC gains if non-natural tree cover (e.g. Oil palm or Eucalyptus plantations) is established on previously cleared areas in the natural forest land use polygons, and (4) potential underestimation of NFTC gains if natural forest gains occur outside the natural forest land use polygons. Uncertainties related to (1) and (2) are inherent to the AGTC data set (and not the proposed method), so they can only be overcome by improving the AGTC automated mapping methodology. Uncertainties (3) and (4) are due to the low temporal resolution of the national land use map, so to minimize these uncertainties and allow for better future forest change estimation, the national land use map would need to be updated periodically (e.g. every 10e15 years). For example, NFTC gains detected
B.A. Johnson / Applied Geography 62 (2015) 294e300
within “natural forest” land use polygons could be removed if an updated forest type map showed that the tree cover gains were due to Oil palm planting rather than natural forest regeneration. On the other hand, if an updated forest type map shows a newlyestablished natural forest, NFTC gains within this new forested area could be included for the historical trend analysis. An additional benefit of updating the land use map would be that the processes driving LULCC in each area of forest loss could be identified (e.g. conversion of forest to agriculture, urban expansion). 5. Conclusions A single-date national forest type map from the year 2000 was integrated with annual global tree cover (AGTC) maps to monitor the annual changes in Thailand's natural forest cover from 2000 to 2012. Thailand's natural forest losses were found to have increased at a near-linear rate over the period, reaching 39,290 ha/year by 2012, and natural forest losses are predicted to increase to 51,775 ha/year by the end of 2015 (an increase of 31.8%) and 66,945 ha/year by 2020 (an increase of 70.4%) if no additional countermeasures are taken. Based on analysis of the long- and short-term trends in annual forest loss by forest type, the forest types with the highest risks of future deforestation/forest degradation (DFD) were identified as Mixed Deciduous, Dry Dipterocarp, and Inundated. Since Mixed Deciduous and Dry Dipterocarp are two of the most dominant forest types in Thailand, reducing the DFD of these forest types is critical to prevent natural forest losses from reaching the predicted levels. The AGTC maps used in this study are planned to be updated annually in the future, and another annual global mapping effort (“forest/non-forest” maps) with similar spatial resolution (25 m) has been started by the Japan Aerospace Exploration Agency (Shimada et al., 2014) (presently limited to the 2007e2010 period, but also planned for updating). These annual tree cover/forest cover maps, when integrated with more detailed (but less frequently updated) national- or sub-national level forest type data sets, have great potential for assessing DFD trends to enable better decisionmaking on how to achieve forest conservation goals related to climate change, biodiversity, and ecosystem services, and the methods presented in this study can serve as a template for other countries' implementation of spatially- and temporally-consistent forest monitoring. Finally, in cases where national data sets with more detailed and/or more accurate forest type information are not available, high resolution (~30 m) global land use/land cover maps containing more general forest information, e.g. the GlobeLand30 maps for the years 2000 and 2010 (Chen et al., 2015), can alternatively be integrated with AGTC maps to assess annual trends in forest change. Acknowledgments This paper is generally based upon outputs produced through projects on “monitoring, reporting, and verification” and “new market mechanisms”, funded by the Japanese Ministry of Environment. References Alix-Garcia, J., De Janvry, A., & Sadoulet, E. (2008). The role of deforestation risk and calibrated compensation in designing payments for environmental services. Environment and Development Economics, 13(3), 375e394. Bellot, F., Bertram, M., Navratil, P., Siegert, F., & Dotzauer, H. (2014). The highresolution global map of 21st-century forest cover change from the University of Maryland (“Hansen Map”) is hugely overestimating deforestation in Indonesia. Retrieved from http://www.forclime.org/documents/press_release/ FORCLIME_Overestimation of Deforestation.pdf. €, M., Norman, A., Lundberg, J., & Elmqvist, T. (2006). The value of Bodin, O., Tengo
299
small size: loss of forest patches and ecological thresholds in southern Madagascar. Ecological Applications, 16(2), 440e451. Bontemps, S., Defourny, P., Bogaert, E., Van, Arino, O., Kalogirou, V., & Perez, J. R. (2011). Globcover 2009: Products description and validation report (pp. 1e53). Retrieved from http://due.esrin.esa.int/globcover/LandCover2009/GLOBCOVER2009_Validation_ Report_2.2.pdf. Chen, J., Chen, J., Liao, A., Cao, X., Chen, L., Chen, X., et al. (2015). Global land cover mapping at 30m resolution: a POK-based operational approach. ISPRS Journal of Photogrammetry and Remote Sensing, 103, 7e27. http://dx.doi.org/10.1016/ j.isprsjprs.2014.09.002. Clinton, N., Yu, L., & Gong, P. (2015). Geographic stacking: decision fusion to increase global land cover map accuracy. ISPRS Journal of Photogrammetry and Remote Sensing, 103, 57e65. http://dx.doi.org/10.1016/j.isprsjprs.2015.02.010. Echeverria, C., Coomes, D. A, Hall, M., & Newton, A. C. (2008). Spatially explicit models to analyze forest loss and fragmentation between 1976 and 2020 in southern Chile. Ecological Modelling, 212, 439e449. Food and Agriculture Organization of the United Nations. (2010a). Global forest resources assessment 2010 country report: Thailand. Food and Agriculture Organization of the United Nations. (2010b). Global forest resources assessment 2010: Main report. Forest Carbon Partnership Facility. (2013). Readiness preparation proposal (R-PP) for Thailand. Retrieved from http://www.forestcarbonpartnership.org/sites/fcp/ files/2013/RRevised RePP version 24 Feb 2013.pdf. Friedl, M. A., Sulla-Menashe, D., Tan, B., Schneider, A., Ramankutty, N., Sibley, A., et al. (2010). MODIS collection 5 global land cover: algorithm refinements and characterization of new datasets. Remote Sensing of Environment, 114(1), 168e182. http://dx.doi.org/10.1016/j.rse.2009.08.016. Hansen, M. C., DeFries, R. S., Townshend, J. R. G., Carroll, M., Dimiceli, C., & Sohlberg, R. A. (2003). Global percent tree cover at a spatial resolution of 500 meters: first results of the MODIS vegetation continuous fields algorithm. Earth Interactions, 7, 1e15. Hansen, M. C., Potapov, P. V., Moore, R., Hancher, M., Turubanova, S. A, Tyukavina, A., et al. (2013). High-resolution global maps of 21st-century forest cover change. Science, 342, 850e853. Hansen, M. C., Stehman, S. V., & Potapov, P. V. (2010). Quantification of global gross forest cover loss. Proceedings of the National Academy of Sciences of the United States of America, 107(19), 8650e8655. Huettner, M., Leemans, R., Kok, K., & Ebeling, J. (2009). A comparison of baseline methodologies for “Reducing emissions from deforestation and degradation.” Carbon Balance and Management, 4, 4. Intergovernmental Panel on Climate Change. (2006). Agriculture, forestry, and other land use. In 2006 IPCC guidelines for national greenhouse gas inventories (Vol. 4) http://dx.doi.org/10.1016/j.phrs.2011.03.002. Margono, B. A., Potapov, P. V., Turubanova, S., Stolle, F., & Hansen, M. C. (2014). Primary forest cover loss in Indonesia over 2000e2012. Nature Climate Change, 4, 730e735. Potapov, P., Yaroshenko, A., Turubanova, S., Dubinin, M., Laestadius, L., Thies, C., et al. (2008). Mapping the world's intact forest landscapes by remote sensing. Ecology and Society, 13(2), 51. Roy, P. S., & Tomar, S. (2000). Biodiversity characterization at landscape level using geospatial modelling technique. Biological Conservation, 95(1), 95e109. Schepaschenko, D., McCallum, I., Shvidenko, A., Fritz, S., Kraxner, F., & Obersteiner, M. (2011). A new hybrid land cover dataset for Russia: a methodology for integrating statistics, remote sensing and in situ information. Journal of Land Use Science, 6(4), 245e259. http://dx.doi.org/10.1080/ 1747423X.2010.511681. Schnittler, M., & Stephenson, S. L. (2000). Myxomycete biodiversity in four different forest types in Costa Rica. Mycologia, 92(4), 626e637. Shimada, M., Itoh, T., Motooka, T., Watanabe, M., Shiraishi, T., Thapa, R., et al. (2014). New global forest/non-forest maps from ALOS PALSAR data (2007e2010). Remote Sensing of Environment, 155, 13e31. Sodhi, N. S., Koh, L. P., Brook, B. W., & Ng, P. K. L. (2004). Southeast Asian biodiversity: an impending disaster. Trends in Ecology and Evolution, 19(12), 654e660. Sodhi, N. S., Posa, M. R. C., Lee, T. M., Bickford, D., Koh, L. P., & Brook, B. W. (2010). The state and conservation of Southeast Asian biodiversity. Biodiversity and Conservation, 19, 317e328. Sohngen, B., & Brown, S. (2006). The influence of conversion of forest types on carbon sequestration and other ecosystem services in the South Central United States. Ecological Economics, 57, 698e708. Song, X.-P., Huang, C., Feng, M., Sexton, J. O., Channan, S., & Townshend, J. R. (2014). Integrating global land cover products for improved forest cover characterization: an application in North America. International Journal of Digital Earth, 7(9), 709e724. http://dx.doi.org/10.1080/17538947.2013.856959. The Royal Forestry Department of Thailand. (2013). Forestry statistics data 2013. Tropek, R. (2014). Comment on “High-resolution global maps of 21st-century forest cover change. Science, 344, 981. http://dx.doi.org/10.1126/ science.1248753. United Nations Framework Convention on Climate Change. (2009). Decision 4/CP.15. United States Department of Agriculture. (1986). Urban hydrology for small watersheds. In Technical release (Vol. 55). Wang, J., Zhao, Y., Li, C., Yu, L., Liu, D., & Gong, P. (2014). Mapping global land cover in 2001 and 2010 with spatial-temporal consistency at 250m resolution. ISPRS Journal of Photogrammetry and Remote Sensing, 103, 38e47. http://dx.doi.org/ 10.1016/j.isprsjprs.2014.03.007.
300
B.A. Johnson / Applied Geography 62 (2015) 294e300
Wichawutipong, J. (2006). Forest area monitoring in Thailand with the use of satellite imagery. In 49th session of the committee on the peaceful uses of outer space. Vienna, Austria. Retrieved from http://www.unoosa.org/pdf/pres/copuos2006/ symposium/06.pdf.
Wohlfart, C., Wegmann, M., & Leimgruber, P. (2014). Mapping threatened dry deciduous dipterocarp forest in South-east Asia for conservation management. Tropical Conservation Science, 7(4), 597e613.