The migration of vegetation under the influence of climate change is of great interest to ecologists, but can be difficult to quantify—especially in less accessible landscapes. Monitoring land cover change through remote sensing has become the best solution, especially with the use of unmanned aerial systems (UASs; drones) as low-cost remote sensing platforms are able to collect data at high spatial and spectral resolutions. Unfortunately, in the context of climate change studies, the lack of comparative UAS data sets over decadal timescales has been limiting. Here, we describe a technique for the integration of historical, low spatial resolution satellite-based Normalized Difference Vegetation Index (NDVI) data with short-term high-resolution multispectral UAS data to track the vegetation changes in a Costa Rican rainforest over a 33-year time frame. The study reveals the transition of a mixed forest from strongly deciduous to weakly deciduous phenology in the Hacienda Barú National Wildlife Refuge (HBNWR), southwestern Costa Rica. This case study presents an approach for researchers and forest managers to study and track vegetation changes over time in locations that lack detailed historical vegetation data. Vegetation migration due to climate change is not well documented and difficult to monitor, especially in remote or inaccessible locations. This case study presents researchers, students, and forest managers an approach for leveraging freely available satellite imagery and UASs to track these changes over time.

## INTRODUCTION

It is widely recognized that climate change is causing significant shifts in temperature and precipitation which alter the regional ecology and distribution of species [1]. Although there is a high potential for vegetation migration in ecosystems, this has not been well documented in inaccessible areas [25]. Many tree species may have a capacity for climate-induced migration, but individual species migrate at markedly different rates [6] leading to changes in tree community composition. Some trees also have the ability to transition their phenological cycles in response to climate change, although a continuing change to climate will favor certain species [7].

Remote sensing has become an essential tool to monitor land cover change in large inaccessible areas [810]. The Normalized Difference Vegetation Index (NDVI) has been shown to be effective in monitoring vegetation’s vigor by measuring the amount of active chlorophyll (red light absorbance) to vegetation cell structure near-infrared reflectance (NIR) as shown in Equation (1) [11].

$NDVI=NIR−RedNIR+Red$
(1)

This monitoring of vegetation’s vigor has also been linked to responses to a change in climate [8, 12, 13]. Monitoring changes in forest also plays a significant role when trying to mitigate climate change [9]. An increasing number of studies have employed the use of time series analysis of all available observations using individual pixels to characterize land cover types [14] and to quantify abrupt changes such as forest clearing [9, 1415]. More recently, by analyzing every Landsat image collected from a study area, interannual and intraannual trajectories and harmonics can be derived from the natural ecosystem to detect very small changes in the environment [16] with the goal of providing a more continuous view of ecosystem dynamics.

In addition to traditionally collected satellite imagery, unmanned aerial systems (UASs) are emerging as a remote sensing platform in ecological applications collecting data at high spatial and spectral resolutions [17]. Terrestrial collection of ground reference data, such as the locations of specific tree species, is challenging tropical locations due to both their inaccessibility and the degradation of the global positioning system (GPS) signal for handheld GPS units under jungle canopy. In response, researchers have demonstrated the ability of UAS to quickly and accurately collect land cover samples in tropical areas for environmental research [18]. These UASs are often equipped with multispectral sensors such as the moderately priced (~\$3,500 in 2017) MicaSense Sequoia (micasense.com/sequoia) which collects on the red edge (735 nm) and near-infrared (790 nm) wavelengths, as well as traditional green (550 nm) and red (660 nm) channels to improve vegetation characterization and classification. UASs are especially well suited for very fine spatial mapping of tropical dry vegetation because imagery can be collected throughout the year enabling the monitoring of seasonal changes in the vegetation. More generally, the accessibility of drone-based imagery has become increasingly important in promoting low-cost, community-based remote sensing project in the tropics [19, 20]

In response to the importance and challenges of monitoring the composition and biodiversity of tropical ecosystems, we present an approach utilizing dense Landsat time series analysis to reveal patterns of phenological change to deciduous trees in a selected forest in southwestern Costa Rica: the Hacienda Barú National Wildlife Refuge (HBNWR). The HBNWR straddles an ecozone between the Lowland Evergreen Moist Forest of Corcovado National Park with less than 35 days of water deficit per annum [21] and the Northern Pacific Lowland Seasonal Forest of Guanacaste and Santa Rosa National Park, which experiences a pronounced dry season of 3 months or more [22]. A 37-year record of rainfall for the HBNWR (http://costarica.jsd.claremont.edu/weather.shtml) suggests that the area may be experiencing a long-term decrease in annual precipitation, presumed to correlate with regional patterns of climate change.

A common component of the HBNWR forest is the Guanacaste tree (Enterolobium cyclocarpum), a semi-deciduous, conspicuous canopy species widely distributed from northern South America through Central America. E. cyclocarpum tolerates significant seasonal drought, as is typical of northwestern Costa Rica, by shedding its leaves. In areas with a less pronounced dry season, E. cyclocarpum may retain some or all of its foliage. The tree becomes progressively rarer in less drought-stressed biomes and is essentially absent in extreme southwestern Costa Rica, except where associated with human disturbance and introduction [23]. Long-term changes in dry season length and severity along a north-south transect of Costa Rica’s Pacific coast can thus be expected to influence E. cyclocarpum success. Moreover, because the tree can respond to drought severity with varying degrees of deciduosity, it has potential as a remotely sensed environmental monitor.

By analyzing individual pixels across all available Landsat images taken from 1984 to 2017 and leveraging fixed-wing UAS platforms equipped with multispectral sensors designed for high taxonomic precision of vegetation species, the temporal pattern of phenological change to deciduous trees in Costa Rica’s mixed forests can be resolved.

## CASE EXAMINATION

### Site Selection and Source Data

In the 1980s, Costa Rica was one of the countries with highest deforestation rates due to the changes for farmland and pasture. According to a study done by the University of Costa Rica’s Center for Tropical Studies and the World Resources Institute, the deforestation rate went from about 46,500 ha/year from 1950 to 1962 to 31,830 ha/year from 1974 to 1989 [24]. A species-rich country, Costa Rica, contains approximately 5% of the Earth’s species making it the most species-dense country in the world, a feat made possible impart because it contains 12 life zones and six climatic transitions [25].

Within Costa Rica, three areas were chosen as the sites of research: 1), a forest on Costa Rica’s central Pacific coast typically subject to an annual drought of approximately 3 months and currently undergoing potential phenological change—the HBNWR; 2) an exemplar of a monsoonal (tropical dry) forest in northwestern Pacific Costa Rica forest, subject to a severe annual drought of 5+ months and hosting a largely deciduous tree community—Santa Rosa National Park; and 3) an exemplar of a tropical wet Costa Rican forest experiencing 12 rainfall months a year and a fully evergreen tree community—Corcovado National Park (Figure 1). These forests are distributed along a north-south transect of Pacific coastal, lowland Costa Rica, and have benefited from being protected areas since before the study period began in 1984.

FIGURE 1.

The study area consists of northern deciduous forests in Santa Rosa National Park, mixed deciduous/tropical forest of the Hacienda Barú, and the tropical wet forests in the southern Corcovado National Park.

FIGURE 1.

The study area consists of northern deciduous forests in Santa Rosa National Park, mixed deciduous/tropical forest of the Hacienda Barú, and the tropical wet forests in the southern Corcovado National Park.

#### Hacienda Barú Refuge

The Hacienda Barú Refuge is a national wildlife refuge and private reserve located in Puntarenas province, Southwestern Costa Rica (9°15′N 84°14′W) (Hacienda Barú 2017). The HBNWR has a drier climate than Corcovado National Park, notably experiencing 3–4 months of marked drought between December and April (http://costarica.jsd.claremont.edu/weather.shtml, Climate Change Knowledge Bank [26]). The Refuge consists of 330 ha having different habitats including primary rainforest, secondary forest, and wetland. Its mission is to preserve and educate others on the natural resources and biodiversity of the area forming one component of the regionally important 82,000 ha Path of the Tapir project [27]. Within this refuge, two regions of interest (ROIs) were selected to monitor changes in vegetation over time. Consisting of 516 Landsat pixels, or an area of 464,029 m2, this area is a mix of tropical and deciduous trees in a fully canopied forest (Figure 2).

FIGURE 2.

The Hacienda Barú study site is a mixed deciduous and tropical, fully canopied forest. Some individual tree crowns can be detected in a high-resolution WorldView-3 satellite image (a) in 25-cm spatial resolution, while individual 30-m pixels are evident in the Landsat image (b).

FIGURE 2.

The Hacienda Barú study site is a mixed deciduous and tropical, fully canopied forest. Some individual tree crowns can be detected in a high-resolution WorldView-3 satellite image (a) in 25-cm spatial resolution, while individual 30-m pixels are evident in the Landsat image (b).

#### Santa Rosa National Park

Santa Rosa National Park (Santa Rosa) is located in Guanacaste province, northwestern Costa Rica (10°84′N, 85°62′W) [28]. Santa Rosa became the first national park in Costa Rica, created in 1971. Santa Rosa has an area of 856.05 km2 and is recognized as a World Heritage Site by UNESCO conserving one of the only areas of tropical dry forest (Holdridge life Zone df-T) in Mesoamerica. Santa Rosa is now one of the few parks which has maintained its land cover with a deforestation rate of 0% [29]. Within this park, two ROIs were selected as exemplars of a protected deciduous forest in western Costa Rica. ROI #1 covered 56,081 m2 or 70 pixels (Figure 3a), and ROI #2 consisted of 38 pixels or an area of 27,188 m2 (Figure 3b).

FIGURE 3.

Two study locations in Santa Rosa National Park were chosen consisting of ROI #1 (32 pixel area, 28,893 m2) (a, b) and ROI #2 (38 pixels, 27,188 m2) (c, d). ROIs are shown in both high-resolution WorldView-3 satellite imagery (a, c) and Landsat imagery (b, d).

FIGURE 3.

Two study locations in Santa Rosa National Park were chosen consisting of ROI #1 (32 pixel area, 28,893 m2) (a, b) and ROI #2 (38 pixels, 27,188 m2) (c, d). ROIs are shown in both high-resolution WorldView-3 satellite imagery (a, c) and Landsat imagery (b, d).

#### Corcovado National Park

Corcovado National Park (Corcovado) is located in the Osa Peninsula in Puntarenas province, Southwestern Costa Rica (8°33′N 83°35′W). It was created as a National Park in 1975 to help preserve the wet tropical forest in the area. Corcovado has a mean annual temperature of 26.9°C and a mean annual precipitation of 3,500 mm in the coast near the study area [30]. Corcovado lies within the tropical wet forest Holdridge life zone, but without a marked dry season. Corcovado is one of the most affected areas by deforestation compared with other national parks or biological reserves [29], although some locations, such as those chosen for this study, have remained as persistent tropical forest.

In Corcovado, two ROIs were identified as tropical wet forest exemplars totaling 154,362.62 m2 or 185 pixels in the satellite imagery (Figure 4). These areas consist of fully canopied tropical wet, evergreen forest, protected, and preserved since 1975.

FIGURE 4.

Two ROIs in Corcovado National Park were chosen consisting of ROI #1 (97 pixels, 85,403 m2) (a, b) and an ROI #2 (88 pixels, 68,958 m2) (c, d). Study areas are shown in both high-resolution WorldView-3 satellite imagery (a, c) and Landsat imagery (b, d).

FIGURE 4.

Two ROIs in Corcovado National Park were chosen consisting of ROI #1 (97 pixels, 85,403 m2) (a, b) and an ROI #2 (88 pixels, 68,958 m2) (c, d). Study areas are shown in both high-resolution WorldView-3 satellite imagery (a, c) and Landsat imagery (b, d).

### Methods

#### Satellite-derived Data

The analysis of the migration of tropical forests was conducted using data sourced from level 1 terrain-corrected (L1T) images, Landsat images. L1T images are georeferenced, terrain corrected, and calibrated across the Landsat constellation of sensors allowing for a direct comparison of pixels over time. The images are prepared and stored by the US Geological Survey Earth Resources Observation and Science (EROS) Science Processing Architecture (ESPA)[9]. Processing options selected included generating surface reflectance, producing NDVI, and using nearest neighbor to preserve individual pixel values. The imagery processing options also included the pixel quality assessment (QA) band containing integers representing atmospheric, instrumental, or cloud contamination factors easily allowing the removal of contaminated pixels. A total of 1,777 Landsat images were analyzed for this study from 1984 to 2017. To automatize data processing, we utilized Python coding to unzip the imagery, composite the bands by date, screen pixels by QA band, and then extract individual pixel values within our delineated study sites. A notional guide to this process is provided in  Appendix A.

#### UAS-derived Data

Two missions of an eBee SenseFly UAS (www.sensefly.com/ebee) were flown on May 31, 2017, at 120 m above ground level (AGL) and third mission flown on June 1 at 600 m AGL (Figure 5). Only the Hacienda Barú Refuge was flown because of the need for higher spatial resolution imagery to differentiate deciduous versus tropical wet tree stands. In Santa Rosa and Corcovado, the tree species is homogeneous, so UAS flights were not necessary. After processing the first mission, the results showed a large section of the study area without images in the processed orthomosaic. This lack of imagery was attributed to the low elevation used in the flight plan and the homogeneous land cover preventing the creation of tie points to mosaic the images. A second flight was needed in order to solve this issue. The eBee specifications suggest that the maximum flight time is 40 min. Taking this into account, the flight plans were designed to have a maximum flight time of 32 min. Identifying a suitable landing and starting point was difficult, with a narrow beach approximately 2.5 km from the center of the study area finally selected. Line of sight was not able to be maintained with the UAS due to the long distance to the study area and thick forest canopy adjacent to the beach. In both missions, a MicaSense Sequoia camera (www.micssense.com/sequoia) was used to provide multispectral images as well as 16 megapixel RGB camera. The flights were planned using eMotion 2, and the data were then processed in Pix4Dmapper.

FIGURE 5.

A large area was flown with a multispectral sensor mounted on a fixed-wing UAS. Multispectral, as well as red, green, blue imagery allowed the team to select study areas that contained both tropical and deciduous vegetation. Image is displayed with near-infrared as red, red edge as green, and green as blue to facilitate vegetation differentiation. UAS, unmanned aerial system.

FIGURE 5.

A large area was flown with a multispectral sensor mounted on a fixed-wing UAS. Multispectral, as well as red, green, blue imagery allowed the team to select study areas that contained both tropical and deciduous vegetation. Image is displayed with near-infrared as red, red edge as green, and green as blue to facilitate vegetation differentiation. UAS, unmanned aerial system.

### Analytical Approach

Analyzing these phenological changes overtime has been shown to be useful to identify climatic variations. Our approach built on this research by creating phenological profiles of deciduous and tropical wet tree stands over the course of the year (intraanual; Figures 6 and 7), shown to be useful in classifying forest composition [31]. To do this, we averaged all NDVI values for each ROI in Santa Rosa (deciduous) and Corcovado (tropical). Individual pixels in the Hacienda Barú Reserve were then selected using UAS imagery and compared with modeled intraanual curves of deciduous exemplars.

FIGURE 6.

Santa Rosa ROI #1 (top) and ROI #2 (bottom) show strong deciduous curves (intraannual) as well as increasing NDVI values across the years (interannual) likely representing the wetter conditions in Costa Rica over the past 10 years. NDVI, Normalized Difference Vegetation Index; ROI, regions of interest.

FIGURE 6.

Santa Rosa ROI #1 (top) and ROI #2 (bottom) show strong deciduous curves (intraannual) as well as increasing NDVI values across the years (interannual) likely representing the wetter conditions in Costa Rica over the past 10 years. NDVI, Normalized Difference Vegetation Index; ROI, regions of interest.

FIGURE 7.

Corcovado ROI #1 (top) and ROI #2 (bottom) show a continual high NDVI average across the seasons. In contrast to the Santa Rosa deciduous forest study area, this tropical area exhibits very little no senescence during the forest 100 days. By breaking up the dates into approximately decadal periods, we do see an increasing NDVI across the days of the year, likely representing the wetter conditions in Corcovado for the past 12 years. NDVI, Normalized Difference Vegetation Index; ROI, regions of interest.

FIGURE 7.

Corcovado ROI #1 (top) and ROI #2 (bottom) show a continual high NDVI average across the seasons. In contrast to the Santa Rosa deciduous forest study area, this tropical area exhibits very little no senescence during the forest 100 days. By breaking up the dates into approximately decadal periods, we do see an increasing NDVI across the days of the year, likely representing the wetter conditions in Corcovado for the past 12 years. NDVI, Normalized Difference Vegetation Index; ROI, regions of interest.

#### Deciduous Phenological Curve (Santa Rosa)

Intraannual NDVI curves in deciduous forest tend to clearly represent a period of senescence, occurring during the first 100 days of the year in the Santa Rosa region [32]. Following this period, rains bring new leaf growth and sustained vigor through the remainder of the calendar year. In Santa Rosa, we averaged 429 images from ROI #1 (Figure 3, top) and 482 from ROI #2 (Figure 3, bottom) to create an intraannual graph to model this phenological cycle (Figure 6). The graphs were created using 10-year temporal brackets to compare the phenological changes over time. Although the intraannual NDVI curve is clear, there is also an interannual (across the years) trend of increased vigor, represented by higher NDVI values over the past 12 years (shown in green) resulting in increasing NDVI values per bracket and showing the past 12 years with even higher NDVI values.

#### Tropical Evergreen Phenological Curve (Corcovado)

The tropical Corcovado study area exhibits a consistently high level of NDVI across the season (Figure 7). This is in contrast to the Santa Rosa deciduous study area, which displays a significant reduction in NDVI for the first 100 days followed by a large increase when the rains return to the area. In Corcovado, we averaged 294 images from ROI #1 (Figure 4, top) and 289 from ROI #2 (Figure 4, bottom). This represents a smaller number of total images than Santa Rosa’s 429 images due to the increased cloud cover over the tropical area. ROI #2 contains less imagery than ROI #1 because on several dates the entire ROI was cloud covered.

Although Santa Rosa has a long, severe annual drought and is just south of the Premonte wet forest, Basal Transition HLZ, Corcovado, is tropical evergreen. In between is the Hacienda Barú’s forest containing mixed deciduous and tropical vegetation. In our analysis, we were interested in how parts or all of the Hacienda Barú’s forest has been changing over time in response to changes in climate. To examine this, we looked at the UAS imagery identifying individual pixels in Hacienda Barú, classifying them as deciduous or tropical evergreen and then comparing them to what we are seeing in Santa Rosa and Corcovado. The analysis of the resulted data mining included averaging the NDVI values per date of image acquisition.

#### Mixed Phenological Curve (Hacienda Barú Refuge)

Although the Santa Rosa and Corcovado study sites represent exemplars of deciduous and tropical forests in western, costal Costa Rica, respectively, the Hacienda Barú Refuge is a mixed forest containing both tropical and deciduous vegetation. To analyze how the vegetation is changing over time, we leveraged UAS multispectral imagery to identify individual pixels within the Hacienda Barú Refuge which were at least 75% contained by either all tropical evergreen or deciduous vegetation for that 30 m2 pixel (Figure 8). From this, we visually identified a sampling of 20 deciduous forest pixels and 20 tropical forest pixels spread over the study area (Figure 9). We then created individual intraannual graphs of these 40 pixels over the 417 Landsat images that were available over this region from 1984 to 2017.

FIGURE 8.

Forty individual pixels (red) were selected for analysis as deciduous or tropical evergreen if they contained at least 75% of their area as one type of vegetation.

FIGURE 8.

Forty individual pixels (red) were selected for analysis as deciduous or tropical evergreen if they contained at least 75% of their area as one type of vegetation.

FIGURE 9.

UAS multispectral imagery over ROI #1 (top) and ROI #2 (bottom) in the Hacienda Barú Refuge was used to manually identify 20 deciduous forest pixels (blue) and 20 tropical forest pixels (red). Image is displayed with near-infrared as red, red edge as green, and green as blue to facilitate vegetation differentiation.

FIGURE 9.

UAS multispectral imagery over ROI #1 (top) and ROI #2 (bottom) in the Hacienda Barú Refuge was used to manually identify 20 deciduous forest pixels (blue) and 20 tropical forest pixels (red). Image is displayed with near-infrared as red, red edge as green, and green as blue to facilitate vegetation differentiation.

When examining the time series NDVI of pixels identified as containing all tropical evergreen vegetation (Figure 10), we see consistently high NDVI values similar to those of a tropical forest (Figure 7). In contrast, when analyzing the pixels identified as deciduous vegetation by approximately 10-year brackets, the transition from a deciduous phenological curve such as in Santa Rosa (Figure 6) to a tropical wet phenological curve becomes evident (Figure 11).

FIGURE 10.

Individual Hacienda Barú Refuge pixels identified through UAS imagery as tropical vegetation such as pixel #75 (top) or pixel #90 (bottom) display a phenological curve typical for a tropical wet evergreen forest such as Corcovado (Figure 7). UAS, unmanned aerial system.

FIGURE 10.

Individual Hacienda Barú Refuge pixels identified through UAS imagery as tropical vegetation such as pixel #75 (top) or pixel #90 (bottom) display a phenological curve typical for a tropical wet evergreen forest such as Corcovado (Figure 7). UAS, unmanned aerial system.

FIGURE 11.

When analyzing individual Hacienda Barú pixels identified through UAS imagery as deciduous such as pixel #72 (top) or pixel #91 (bottom), we see a typical deciduous forest phenological curve from 1984 to 1992, but a transition to a tropical evergreen phenological curve by 2005 to present. UAS, unmanned aerial system.

FIGURE 11.

When analyzing individual Hacienda Barú pixels identified through UAS imagery as deciduous such as pixel #72 (top) or pixel #91 (bottom), we see a typical deciduous forest phenological curve from 1984 to 1992, but a transition to a tropical evergreen phenological curve by 2005 to present. UAS, unmanned aerial system.

## RESULTS

To quantify the transition of Hacienda Barú Refuge from a semi-deciduous or mixed forest to one of the tropical wet characteristics, we calculated the average vegetation vigor and standard deviation: 20 deciduous and 20 tropical wet pixels from the Hacienda Barú Refuge at roughly decadal periods (Table 1). Our results show that from 1984 to 1994, deciduous pixels more closely matched the deciduous phenological curve from Santa Rosa with lower average NDVI (0.74) and a higher standard deviation (0.079). But by the last decadal period, the pixels more closely resembled the tropical exemplar with an NDVI (0.873) very similar to those pixels in Corcovado or those identified as tropical in the Refuge (0.874). Their annual deviation also reduced showing a significant reduction in dry season senescence, more closely the tropical wet phenological curve exemplar of Corcovado by the end of the study period.

TABLE 1.

The phenological curve of areas identified as deciduous in the Hacienda Barú Refuge tended to match the curve of the deciduous Santa Rosa forest for the first time period (1984–1994). However, by the last time period, these areas more closely resembled the curve of the tropical Corcovado forest. In contrast, areas identified in the refuge as tropical, always best fit the phenological curve of the tropical Corcovado

IDType1984–19941995–20042005–2017
AverageStandard deviationAverageStandard deviationAverageStandard deviation
27 Deciduous 7,712 959 8,248 731 8,715 480
58 Deciduous 8,106 558 8,559 408 8,869 388
67 Deciduous 7,980 602 8,393 536 8,702 458
72 Deciduous 6,651 1,180 8,424 486 8,702 527
73 Deciduous 6,365 1,296 8,271 648 8,681 796
91 Deciduous 7,181 978 8,483 431 8,687 512
92 Deciduous 6,629 1,317 8,514 494 8,708 806
128 Deciduous 7,004 845 8,124 581 8,560 405
136 Deciduous 8,024 556 8,520 488 8,814 365
145 Deciduous 6,553 1,052 8,133 606 8,599 577
171 Deciduous 7,085 664 8,166 584 8,667 414
184 Deciduous 7,340 614 8,155 577 8,737 428
194 Deciduous 7,837 569 8,487 622 8,978 333
224 Deciduous 8,038 576 8,392 547 8,777 431
229 Deciduous 7,915 556 8,303 518 8,743 343
245 Deciduous 8,037 597 8,616 519 9,032 283
287 Deciduous 7,371 610 8,404 506 8,753 363
322 Deciduous 6,905 1,094 8,462 507 8,779 346
369 Deciduous 7,804 695 8,257 397 8,659 353
447 Deciduous 7,766 640 8,204 697 8,672 434

Average  7,397 806 8,364 536 8,745 453

Tropical 7,855 696 8,189 515 8,616 460
Tropical 7,924 558 8,200 569 8,673 451
41 Tropical 8,000 609 8,516 428 8,666 411
57 Tropical 8,219 624 8,472 468 8,826 329
75 Tropical 8,054 655 8,369 529 8,639 516
87 Tropical 7,968 612 8,425 477 8,785 413
90 Tropical 7,793 599 8,471 400 8,725 493
148 Tropical 7,885 640 8,261 517 8,713 439
186 Tropical 7,943 503 8,202 723 8,708 406
199 Tropical 8,028 446 8,367 635 8,832 364
211 Tropical 8,088 532 8,327 600 8,793 346
234 Tropical 8,111 485 8,482 517 8,864 300
237 Tropical 8,011 525 8,339 527 8,742 379
304 Tropical 7,945 618 8,358 464 8,679 471
365 Tropical 8,030 537 8,517 437 8,811 339
367 Tropical 7,915 477 8,276 453 8,593 346
398 Tropical 8,056 481 8,392 492 8,774 310
455 Tropical 7,890 512 8,349 566 8,707 460
461 Tropical 7,982 570 8,511 487 8,828 380
486 Tropical 7,756 601 8,452 416 8,633 373

Average  7,973 564 8,374 511 8,730 399
IDType1984–19941995–20042005–2017
AverageStandard deviationAverageStandard deviationAverageStandard deviation
27 Deciduous 7,712 959 8,248 731 8,715 480
58 Deciduous 8,106 558 8,559 408 8,869 388
67 Deciduous 7,980 602 8,393 536 8,702 458
72 Deciduous 6,651 1,180 8,424 486 8,702 527
73 Deciduous 6,365 1,296 8,271 648 8,681 796
91 Deciduous 7,181 978 8,483 431 8,687 512
92 Deciduous 6,629 1,317 8,514 494 8,708 806
128 Deciduous 7,004 845 8,124 581 8,560 405
136 Deciduous 8,024 556 8,520 488 8,814 365
145 Deciduous 6,553 1,052 8,133 606 8,599 577
171 Deciduous 7,085 664 8,166 584 8,667 414
184 Deciduous 7,340 614 8,155 577 8,737 428
194 Deciduous 7,837 569 8,487 622 8,978 333
224 Deciduous 8,038 576 8,392 547 8,777 431
229 Deciduous 7,915 556 8,303 518 8,743 343
245 Deciduous 8,037 597 8,616 519 9,032 283
287 Deciduous 7,371 610 8,404 506 8,753 363
322 Deciduous 6,905 1,094 8,462 507 8,779 346
369 Deciduous 7,804 695 8,257 397 8,659 353
447 Deciduous 7,766 640 8,204 697 8,672 434

Average  7,397 806 8,364 536 8,745 453

Tropical 7,855 696 8,189 515 8,616 460
Tropical 7,924 558 8,200 569 8,673 451
41 Tropical 8,000 609 8,516 428 8,666 411
57 Tropical 8,219 624 8,472 468 8,826 329
75 Tropical 8,054 655 8,369 529 8,639 516
87 Tropical 7,968 612 8,425 477 8,785 413
90 Tropical 7,793 599 8,471 400 8,725 493
148 Tropical 7,885 640 8,261 517 8,713 439
186 Tropical 7,943 503 8,202 723 8,708 406
199 Tropical 8,028 446 8,367 635 8,832 364
211 Tropical 8,088 532 8,327 600 8,793 346
234 Tropical 8,111 485 8,482 517 8,864 300
237 Tropical 8,011 525 8,339 527 8,742 379
304 Tropical 7,945 618 8,358 464 8,679 471
365 Tropical 8,030 537 8,517 437 8,811 339
367 Tropical 7,915 477 8,276 453 8,593 346
398 Tropical 8,056 481 8,392 492 8,774 310
455 Tropical 7,890 512 8,349 566 8,707 460
461 Tropical 7,982 570 8,511 487 8,828 380
486 Tropical 7,756 601 8,452 416 8,633 373

Average  7,973 564 8,374 511 8,730 399

## CONCLUSION

This study presents a unique analytical approach for measuring the phenological changes to individual trees in mixed tropical forests. By combining UAS multispectral data from a single set of flights (high spatial resolution, low temporal resolution) with Landsat satellite imagery since 1984 (low spatial resolution, high temporal resolution), we are able to identify and document the transition of Guanacaste trees in a mixed forest from strongly deciduous to weakly deciduous behavior more closely resembling the phenological cycle of tropical evergreen vegetation in the country. These trends were identified by comparing deciduous pixels in the study area with exemplars developed from the deciduous forest of Santa Rosa and the tropical wet forest of Corcovado.

Furthermore, this analysis identifies how this transition of the deciduous trees in the Hacienda Barú Refuge correlates with the wetter conditions mid-latitude Costa Rica has experienced over the years. If the trend of wetter climate continues in mid-latitude Costa Rica, conditions will increasingly favor tropical evergreen vegetation threatening the existence of deciduous trees in these mixed forests. These heterogeneous forests provide a habitat for some of the world’s richest biodiversity and are under threat from changing climate conditions. In response, this manuscript demonstrates a functional approach for the continued monitoring of the phenological processes of vegetation in mixed tropical forests and serves as a warning that the rich biodiversity of Costa Rica’s mixed, mid-latitude forests is under threat from climate change.

## CASE STUDY QUESTIONS

1. What is the connection between global climate change and vegetation migration? Why does geography make it especially hard to examine this connection?

2. How does publicly available, historical satellite imagery serve as a resource for all of us in tracking global climate change? How can this resource be used to project future threats to biodiversity?

3. How can UAS be used in your local community to track and describe changes to the natural environment?

### APPENDIX A

Notional guide to download and processing temporal stack of Landsat imagery.

1. Log onto EarthExplorer.usgs.gov and select the area and date period you are interested in

2. Under the “Data Sets” tab, Select Landsat Analysis Ready Data (ARD) under Landsat

3. Under “Additional Criteria,” select Cloud Cover “Less than 10%”

4. Select desired images for “Bulk Download” (see attached)

6. In Esri ArcMap or ArcGIS Pro, use the tool “Extract by Points” to extract the raster cell values over the coordinated you are interested in. Use Modelbuilder to iterate through a folder where all the Landsat images are.

7. In this manuscript, we extracted red and near Infrared, and in excel calculated Normalized Difference Vegetation Index (NDVI).

8. In Excel, the “YEAR” and “DATE” functions can be used to convert to Julian. This allows for the creation of Figures 6 and 7 in this paper.

## AUTHOR CONTRIBUTIONS

AM contributed to research design, analysis, and manuscript creation. DM assisted in the analysis and manuscript creation.

We would like to thank Clayton Sayre and Warren Roberts for their contributions to field data collection. We would also like to thank the Claremont Graduate University for their support for this research.

## FUNDING

The BLAIS Foundation scholarship titled “Connecting Imagery-Based Data Science with Environmental Analysis: Collaborative Fieldwork in Costa Rica and Analysis in Claremont” (Grant number: 221-2170058) supported Andrew Marx.

## COMPETING INTERESTS

The authors have declared that no competing interests exist.

## REFERENCES

REFERENCES
1.
IPCC
. IPCC Second Assessment Climate Change 1995. Intergovernmental Panel on Climate Change.
1995
. Available: https://www.ipcc.ch/pdf/climate-changes-1995/ipcc-2nd-assessment/2nd-assessment-en.pdf
2.
Walther GR, Post E, Convey P, Menzel A, Permesan, C, Beebee T, . . . Bairlein F
Ecological responses to recent climate change
.
Nature.
2002;
416
. Available: https://www.nature.com/nature/journal/v416/n6879/pdf/416389a.pdf
3.
Caplat P, Anand M, Bauch C.
Interactions between climate change, competition, dispersal, and disturbances in a tree migration model
.
Theor Ecol
.
2008
;
1
(
4
):
209
220
.
4.
Malcolm JR, Markham A, Neilson RP, Garaci M.
Estimated migration rates under scenarios of global climate change
.
J Biogeogr
.
2002
;
29
(
7
):
835
849
.
5.
Kelly A, Goulden M.
Rapid shifts in plant distribution with recent climate change
.
Proc Natl Acad Sci U S A
.
2008
;
105
(
33
):
11823
11826
.
6.
King GA, Herstrom AA.
Holocene tree migration rates objectively determined from fossil pollen data
. In: Huntley B, Cramer W, Morgan AV, Prentice HC, Allen JRM, editors.
Past and Future Rapid Environmental Changes. NATO ASI Series (Series I: Global Environmental Change)
. Vol.
47
.
Berlin, Heidelberg
:
Springer
;
1997
. pp.
91
101.
7.
Hughen KA, Eglinton TI, Xu L, Makou M.
Abrupt tropical vegetation response to rapid climate changes
.
Science
.
2004
;
304
(
5679
):
1955
1959
.
8.
Hayes DJ, Sader SA.
Comparison of change-detection techniques for monitoring tropical forest clearing and vegetation regrowth in a time series
.
Photogram Eng Remote Sens
.
2001
;
67
(
9
):
1067
1075
.
9.
DeVries B, Decuyper M, Verbesselt J, Zeileis A, Herold M, Joseph S.
Tracking disturbance-regrowth dynamics in tropical forests using structural change detection and Landsat time series
.
Remote Sens Environ
.
2015
;
169
:
320
334
.
10.
Coppin P, Jonckheere I, Nackaerts K, Muys B, Lambin E.
Digital change detection methods in ecosystem monitoring: a review
.
Int J Remote Sens
.
2010
;
25
(
9
):
1565
1596
.
11.
Tucker CJ.
Red and photographic infrared linear combinations for monitoring vegetation
.
Remote Sens Environ
.
1979
;
8
(
2
):
127
150
.
12.
Brown ME, Pinzon JE, Didan K, Morisette JT, Tucker CJ.
Evaluation of the consistency of long-term NDVI time series derived from AVHRR,SPOT-vegetation, SeaWiFS, MODIS, and Landsat ETM+ sensors
.
IEEE Trans Geosci Remote Sens
.
2006
;
44
(
7
):
1787
1793
.
13.
Xu Y, Yang J, Chen Y.
NDVI-based vegetation responses to climate change in an arid area of China
.
Theor Appl Climatol
.
2016
;
126
(
1–2
):
213
222
.
14.
Zhu, Zhe, and Curtis E. Woodcock. “Continuous change detection and classification of land cover using all available Landsat data.”
Remote sensing of Environment
144
(
2014
):
152
171.
15.
Zhu, Zhe, Curtis E. Woodcock, and Pontus Olofsson. “Continuous monitoring of forest disturbance using all available Landsat imagery.”
Remote sensing of environment
122
(
2012
):
75
91
.
16.
Pasquarella VJ, Holden CE, Kaufman L, Woodcock CE.
From imagery to ecology: leveraging time series of all available Landsat observations to map and monitor ecosystem state and dynamics
.
Remote Sens Ecol Conserv
.
2016
;
2
(
3
):
152
170
.
17.
Koh L, Wich S.
Dawn of drone ecology: low-cost autonomous aerial vehicles for conservation
.
Trop Conserv Sci
.
2012
;
5
(
2
):
121
132
.
18.
Marx A, Mcfarlane D, Alzahrani A.
Uav data for multi-temporal Landsat analysis of historic reforestation: a case study in Costa Rica
.
Int J Remote Sens
.
2016
;
38
(
8–10
):
2331
2348
.
19.
Paneque-Gálvez J, Mccall M, Naploetano B, Wich S, And Koh L.
Small drones for community-based forest monitoring: An assessment of their feasibility and potential in tropical areas
.
Forests
.
2014
;
5
(
6
):
1481
1507
.
20.
Radjawali I, Pye O.
Drones for justice: inclusive technology and river-related action research along the Kapuas
.
Geogr Helv
.
2017
;
72
(
1
):
17
.
21.
Gilbert LE, Christen CA, Altricher M et al.
The Southern Pacific Lowland Evergreen Moist Forest of the Osa Region
. In: Kappelle M, editor.
Costa Rican Ecosystems
.
Chicago, Ill
:
The University of Chicago Press
;
2016
. pp.
360
411
.
22.
Quirico Jiminez M, Eduardo Carillo J, Kappelle M.
The Northern Pacific lowland seasonal dry forests of Guanacaste and the Nicoya Peninsula
. In: Kappelle M, editor.
Costa Rican Ecosystems
.
Chicago, Ill
:
The University of Chicago Press
;
2016
. pp.
247
289
.
23.
Janzen DH.
Enterolobium cyclocarpum
. In: Janzen DH, editor.
Costa Rican Natural History
.
Chicago
:
University of Chicago Press
;
1983
. pp.
241
243
.
24.
DeCamino R, Segura O, Arias LG, & Pérez I. Costa Rica Forest Strategy and the Evolution of Land Use. The International Bank for Reconstruction and Development/World Bank.
2000
. Available: http://documents.worldbank.org/curated/en/238731468023661911/Costa-Rica-Forest-strategy-and-the-evolution-of-land-use
25.
Kappelle M, Lovejoy TE, Gámez Lobo R.
Costa Rican Ecosystems
.
Chicago
:
University of Chicago Press
;
2016
.
26.
World Bank Group
. Climate Change Knowledge Portal. Available: http://sdwebx.worldbank.org/climateportal. Accessed Sept
2017
.
27.
Redondo-Brenes A, Chiu A, Snow S.
Small-Scale Restoration Efforts Using Mixed-Tree Plantations in the Path of the Tapir Biological Corridor, Costa Rica
.
Trop Resour Bull
.
2010
;
29
:
27
32
.
28.
Powers JS, Becknell JM, Irving J, Pèrez-Aviles D.
Diversity and structure of regenerating tropical dry forests in Costa Rica: Geographic patterns and environmental drivers
.
For Ecol Manage
.
2009
;
258
(
6
):
959
970
.
29.
Sánchez-Azofeifa G, Daily GC, Pfaff ASP, Busch C.
Integrity and isolation of Costa Rica’s national parks and biological reserves: examining the dynamics of land-cover change
.
Biol Conserv
.
2003
;
109
(
1
):
123
135
.
30.
Salom-Pérez R, Carrillo E, Sáenz JC, Mora JM.
Critical condition of the jaguar Panthera onca population in Corcovado National Park, Costa Rica
.
Oryx
.
2007
;
41
(
1
):
51
56
.
31.
Jimenez Galo AJ.
Monitoring of Tropical Forest Cover with Remote Sensing
. In: Pancel L, Köhl M, editors.
Tropical Forestry Handbook
.
Berlin, Heidelberg
:
Springer
;
2016
. pp.
663
685
. doi:.
32.
Frankie GW, Baker HG, Opler PA.
Comparative phenological studies of trees in tropical wet and dry forests in the lowlands of Costa Rica
.
J Ecol
.
1974
;
62
(
3
):
881
919
.