Forest change within and outside protected areas in the Dominican Republic, 2000-2016

We used Landsat-based estimates of tree cover change to document the loss and gain of forest in the Dominican Republic between 2000 and 2016. Overall, 2,795 km2 of forest were lost, with forest gain occurring on only 393 km2, yielding a net loss of 2,402 km2 of forest, a decline of 11.1% or 0.7% per year. Deforestation occurred in all of the major forest types in the country, and ranged from a 13% decline in the area of semi-moist broadleaf forest to a 5.9% loss of cloud forest, mostly attributed to agriculture. Fire was a significant driver of forest loss only in Hispaniolan pine (Pinus occidentalis) forests and, to a lesser extent, in adjacent cloud forest. Deforestation rates were lower within protected areas, especially in dry and semi-moist broadleaf forests at lower elevations. Protected areas had a smaller, and generally negligible, effect on rates of forest loss in pine forest and cloud forest, largely due to the effects of several large wildfires. Overall, rates of deforestation in the Dominican Republic were higher than regional averages from across the Neotropics and appeared to have accelerated during the later years of our study period. Stemming deforestation will likely require enforcement of prohibitions on large-scale agricultural production within protected areas and development of alternatives to short-cycle, shifting agriculture.


30
Human well-being is linked inextricably with the fate of the planet's forests. Forests 31 provide goods and income to the rural poor throughout the developing world [1], generate 32 employment for more than 10 million people throughout the world [2], yield renewable flows of 33 raw materials for commercial and domestic use, sustain stable flows of clean water [3,4], buffer 34 against local extremes of climate [5], and regulate global climate and carbon cycles [6,7]. 35 Indeed, the very persistence of modern human societies may be incompatible with the 36 conditions created by ongoing deforestation [8]. The survival of an uncounted number of non-37 human species also depends on the persistence of forested landscapes. 38 Efforts to conserve Earth's remaining forests, and to understand the consequences of 39 their disappearance, demand estimates of where, and at what rate, forest loss is occurring 40 [9,10]. Reliable national-level data on forests is urgently needed to inform policies on forest 41 conservation, sustainable development, and climate-change mitigation. A significant contribution 42 to these efforts was made by Hansen et al. [9], who provided satellite-based estimates of global 43 forest cover at a relatively fine temporal and spatial scale. Those data have been used 44 subsequently to generate regional estimates of deforestation [11], estimates of loss of specific 45 forest types [12], and country-specific descriptions of forest change [13]. Although analyses at 46 planetary and regional scales provide useful insights for efforts to limit the deleterious 47 consequences of global change [14] or meet global sustainable development goals [15], 48 smaller-scale analyses, especially at the national or sub-national level, are useful because they 49 align more closely with the level at which policies on forest use and conservation are 50 implemented. Thus, country-specific analyses of deforestation allow for the evaluation of the 51 efficacy of conservation interventions and, ideally, implementation of adaptive changes as 52 needed. 53 Here, we examine spatial and temporal patterns of change in forest cover in the 55 Dominican Republic (DR) between 2000 and 2016 using Hansen et al.'s [9] forest-cover dataset 56 and its annual updates. In particular, we document changes in the extent of forest cover, by 57 forest type, and examine the efficacy of the nation's system of protected areas -the country's 58 primary conservation tool -in stemming forest loss. We focused on the DR for several reasons. 59 First, as a middle-income country, it is broadly reflective of the changing dynamics and 60 challenges faced globally in conserving forests in developing countries experiencing rapid 61 economic growth: the DR's average economic growth of 5.3% over the past  The forest types considered in the second analysis include Hispaniolan pine (Pinus 100 occidentalis) forest, which was classified by Tolentino and Peña [23] into both an open ("bosque 101 conífera abierto") and closed-canopy ("bosque conífera denso") category; cloud forest ("bosque 102 nublado"); moist broadleaf forest ("bosque húmedo"); semi-moist broadleaf forest ("bosque 103 semihúmedo"); and dry forest ("bosque seco" MODIS-based estimates of the global area burned [25]. We aggregated monthly estimates of 119 area burned for each year and assumed that forest loss was caused by fire for any pixel in the 120 Hansen et al.
[9] data that was within the boundaries of a burned area and was estimated as 121 having been deforested in that year.

122
Quantifying forest change within protected areas 123 The DR has an extensive national protected area system, covering 26% of its territory 124 [26]. To examine whether forest within formally protected areas showed different patterns of 125 change, we calculated forest change and area burned for each protected area within the DR.

127
Quantifying forest change 128 Trees covered 21,494 km 2 of the DR in 2000, roughly 45% of its total land area. 129 Deforestation removed 2,795 km 2 of this tree cover by 2016, while reforestation or afforestation occurred on only 393 km 2 , a net loss of 2,402 km 2 , reducing forest cover to roughly 40% of the 131 territory. This amounts to an 11.1% decline in forest cover at the national level over the period of 132 analysis, an annual deforestation rate of 0.7%. 133 Considering only the DR's major natural forest types, forest cover shrank from 9,517 km 2 134 in 2000 to 8,644 km 2 in 2016, a net loss of 9.2% (Table 1). Depending on forest type, this 135 change ranged from -5.9% in cloud forests to -13.1% in semi-moist forests. The extent of loss 136 varied among years but, with the exception of dry forest, tended to increase after 2010 (Fig 1). 137 Forest gain was negligible in all of the natural forest types. Fire as a deforestation driver 150 Fire accounted for a significant amount of loss in the Hispaniolan pine forests, in both 151 open-and closed-canopy types (Table 1). However, most of the loss caused by fire occurred in 152 a single year (Fig 2). Protected area deforestation 168 Rates of deforestation within protected areas largely mirrored overall trends in forest 169 change, except for dry and semi-moist broadleaf forests where forest loss was substantially 170 lower within protected areas (Table 2). Within protected areas, forest accounted for 7,381 km 2 in 171 2000, covering 57% of the land. By 2016, forest cover had shrunk by 670 km 2 (-8.5%) and 172 covered only 52% of the land in protected areas. 173 Protected areas offered little defense against fire, either. Fire accounted for significant 174 amounts of the estimated loss of both pine and cloud forests within protected areas ( Cloud forest losses within protected areas ranged from <1 km 2 in Armando Bermúdez 183 National Park to 11 km 2 in JC Ramírez, or 17% of that park's extant cloud forest ( Fig. 5; S1 184 File). Roughly half (51%) of the cloud forest lost in JC Ramírez was due to the same wildfires 185 that burned through the park's pine forests. Other parks experiencing substantial loss of cloud 186 forest were Valle Nuevo National Park, which lost 7.7 km 2 (4.1% of its extent in 2000), and 187 Bahoruco, which lost 7.6 km 2 (8.2%). Loss of cloud forest in these two parks was driven 188 primarily by processes other than fire (only 26.3% and 8.8%, respectively, of the deforestation in 189 each was caused by fire). 190  Category II) except for Alto Bao, a forest reserve (IUCN Category V), and collectively accounting for 94% of the total protected area for this forest type -was mostly due to sources 217 other than fire. Fire was an important source of deforestation only in José del Carmen Ramírez 218 National Park; Sierra de Bahoruco National Park and Valle Nuevo National Park both lost large 219 areas of cloud forest from causes other than fire. 220 221 Moist broadleaf forest losses were greatest in Los Haitises National Park, which lost 26 222 km 2 (14.6%), almost all (94.2%) due to causes other than fire ( Fig. 6; S1 File) Loss of semi-moist broadleaf forest was most pronounced in Bahoruco (13.9 km 2 , or 236 15.5% of the 2000 total extent) and Cotubanamá National Park (formerly Del Este National 237 Park; 6 km 2 , or 2%; Fig. 7; S1 File). Bahoruco also led all parks in the amount of dry forest 238 eliminated, with 7.8 km 2 (5.1%) lost over the course of this study ( Fig. 8 of which are essential in generating reliable national reports on forest change. Given this, and 287 given the consistency of estimates produced by international studies, we consider it unlikely that 288 reforestation exceeded deforestation and instead have high confidence that the total area of 289 forest in the DR declined from 2000-2016. 290 One possible source of error in our estimates of net deforestation is that our estimates of 291 gain in the area of each forest type apply only to pixels falling within the mapped distribution of 292 each forest type. Because we based our estimates of change in each forest type on its 1996 293 mapped distribution, we cannot rule out the possibility that areas categorized as another land-294 cover type in 1996 (e.g., subsistence agriculture) could have regrown into one of the forest 295 types we analyzed. This would not have been captured by our analysis, thus leading us to 296 underestimate forest gains during the period. However, the total gain in tree cover across all of 297 the agricultural or otherwise anthropogenic land-cover types in the 1996 land-cover map was 298 only 24 km 2 , so even if all of this gain reflected reversion to native forest cover, which is unlikely, 299 it would account for only a small fraction of the 874 km 2 of forest lost. Thus, we are confident that afforestation of agricultural or developed lands could not have materially affected our 301 estimates of net loss. 302 As has been reported in other studies of deforestation in the Neotropics [30,32], we also 303 found that deforestation in the DR tended to accelerate over time, with the exception of dry 304 forest loss, which showed some evidence of a decline in the extent of deforestation after 2010. 305 This slowing deforestation rate in dry forests could be because of the substitution of propane 306 gas for wood charcoal -the main historical use of dry-forest trees -as the primary cooking fuel 307 in

315
Outside of areas known to have burned, the data that we used do not provide direct 316 insight into the drivers of forest loss. However, we can reasonably speculate that, with the 317 exception of pine forests, the most likely cause for the observed forest loss is expanding 318 agriculture. This is not only consistent with our field observations, but also in agreement with the 319 findings from a comprehensive, national-level assessment which ranked agriculture as the 320 leading cause of deforestation, accounting for 55% of forest loss in the DR [37]. In comparison, 321 the same study attributed only 26% of deforestation to timber harvesting, firewood collection, 322 and wood-charcoal production. 323 The important role of agriculture in forest clearing in two montane national parks has 324 also been highlighted in recent reports by Wooding and Morales [38] for Nalga de Maco 325 National Park and León et al. [39] for Sierra de Bahoruco National Park. Both studies describe 326 the expansion of a similar commercial agricultural system, consisting of sharecropping in a 327 shifting-agriculture system established between a landless Haitian farmer and a Dominican who 328 claims land ownership. Sharing arrangements can vary, but usually the farmer keeps most of 329 the crop, which is typically short-cycle crops. León et al. [39] also described the recent 330 establishment of more permanent forest conversion in the form of avocados (Persea americana) 331 grown for export, plantations of which have actively expanded inside Sierra de Bahoruco 332 National Park since 2008. The problem of agriculture within protected areas is not limited to 333 montane parks, however; a study on the drivers of deforestation in the low-elevation Los 334 Haitises National Park also identified farming as the leading cause. In this case, deforestation 335 was driven by increased exports of taro root (Colocasia esculenta), the leading crop inside the 336 Park [40]. 337 Fire was the leading cause of forest-cover decline in Hispaniolan pine forests. Pine trees 338 and their associated understory plants are not only resilient to fire, but depend on it for seed 339 dispersal and germination [41] and thus, absent any additional disturbance, burned pinelands 340 will likely recover [42]. Of concern, however, is evidence of emerging changes in fire regime that 341 may pose a long-term threat to these forests. Whereas lightning during dry seasons was probably the leading cause of fire ignition in the past, today human activities are. The DR's 343 National Fire Management Strategy has identified as the leading causes of forest fires, in order 344 of importance: farming activities (especially land preparation for short-cycle crops), renewal of 345 cattle grazing pastures, intentional fires in protest against authorities, and accidental fires 346 caused by abandoned cooking fires from hunters and parrot poachers [43]. Furthermore, the 347 strategy highlights a new and complex threat: the expansion of the invasive molasses grass 348 (Melinis minutiflora), which is highly flammable and has already been implicated in forest fires 349 [43]. Changes in the seasonality, frequency, or intensity of fire may negatively affect even 350 relatively resilient pine forests, let alone broadleaf forests that are ill-adapted to fire. 351 Cloud forest also experienced substantial losses due to fire. However, unlike pine forest, 352 it is far less resilient to fire. Not only is cloud forest exceedingly slow to recover after fire [44], 353 but exposure to repeated fire can lead to its replacement by other forest types [42]. The fire-354 related losses of cloud forest that we documented, therefore, may be permanent. This is very 355 concerning as these montane forests not only host most of the unique, threatened species on 356 the island, but also intercept water from rain and clouds year-round (e.g., who also found that protected areas appear more secure when established in areas not highly 395 valued for extractive resource uses. The apparently greater effectiveness of protected areas in 396 areas of dry forest in the DR may thus simply reflect the low profitability of exploiting the 397 resources that they contain, in contrast to the relatively lucrative opportunities afforded by the 398 export-oriented agriculture that can be carried out in protected areas with more suitable climatic 399 conditions.

400
Policy implications 401 Although not typically considered a hotspot of deforestation, rates of forest loss in the 402 DR are higher than regional averages and show no sign of decelerating. Our results reveal 403 ongoing deforestation across the country, especially in moist forest types that are more valuable 404 for agricultural development. Protected areas offered only modest reductions in deforestation for 405 most forest types, highlighting a general lack of management effectiveness. As nations continue 406 to expand their protected-area systems, there is an urgent need to undertake objective 407 assessments of their effectiveness in meeting their goals, especially those pertaining to forest 408 conservation. Satellite images and forest-cover analysis platforms, such as Global Forest 409 Watch, offer an inexpensive and objective way to achieve this. 410 Continued deforestation in the DR poses a risk to the flow of critical ecosystem services, 411 especially the provision of water by upland forests to lowland human communities, including the 412 major cities and agricultural regions. Ongoing deforestation will also threaten the achievement 413 of a number of the DR's sustainable development goals, as well as meeting its Intended 414 Nationally Determined Contribution under the Paris Agreement within the United Nations 415 Framework Convention on Climate Change (UNFCCC). Widespread forest loss will also hinder 416 the DR's commitments to halt biodiversity loss as a party to the Convention on Biological 417 Diversity, by placing at greater risk many unique, globally threatened species that depend on 418 the country's forests. 419 Addressing deforestation will require a better understanding of its causes. Although fire 420 is an important driver of loss of forest cover in Hispaniolan pine forest, and occasionally in 421 adjacent cloud forest, the vast majority of deforestation is driven by clearing for agricultural 422 production [53]. More research into the local drivers of deforestation, its key actors, and 423 associated social dynamics are needed. Efforts to stem deforestation will almost certainly 424 involve stricter limits on large-scale agricultural commodity production within protected areas 425 and the development of alternative livelihood opportunities for those practicing shifting 426 agriculture. Shifting agriculture is in great part enabled by customary systems of land tenure in 427 many rural areas of the DR that persist despite contravening laws and policies established by 428 the central government. The critical role of land tenure in reducing deforestation, particularly 429 under the REDD+ (Reducing Emissions from Deforestation and Forest Degradation) 430 mechanism of the UNFCCC has been highlighted by a growing number of studies around the 431 world [e.g., 54,55]. Addressing these issues is not easy, but will be crucial for securing the 432 future of forests in the DR and in many other countries facing similar development pressures.