Thirty-two years of mangrove forest land cover change in Parita Bay, Panama

Abstract Mangrove forests have experienced a rapid decline. However, the rate of loss has decreased in recent years due to enhanced conservation and nature regeneration. The dynamics of mangrove forests in Panama have not been monitored since the year 2000, despite a significant loss during the 1980s. The objectives of our study were to quantify changes in mangrove cover and identify the dominant drivers of change in Parita Bay, Panama. Temporal changes in mangrove cover and the Normalized Difference Vegetation Index (NDVI) were determined using the supervised classification method on Landsat satellite images from 1987 to 2019. We identified a 4.7% increase in the mangrove area of Parita Bay during the 32 years; the mangrove forests were also considered healthy as reflected by high NDVI values. However, the conversion of mangroves to other land cover types resulted in a 1.26% decline in mangrove cover from 1987 to 1998. Moreover, the area of aquaculture and saltpans almost doubled during this period. During the following two decades, the conversion of other land cover classes (water, other vegetation, and bare soil) increased the mangrove area by 6%, and the annual rate of increase was greater during the second decade (0.43% year−1). From 2009 to 2019, mangroves declined at an annual rate of 0.11% in protected areas and increased at an annual rate of 0.50% in unprotected areas. Despite the regeneration potential of mangrove forests, our study highlights the need to continually manage and protect mangrove forests in order to facilitate their expansion in Parita Bay.

This last one, referring to the effect of climatic variables on the mangrove ecosystem is poorly understood. As Climate Change brings an irregular distribution of rainfall worldwide (Solomon et al., 2007), effects over mangrove forest will vary. For example, increasing rainfall may result in a greater mangrove growth rate. While, decreasing rainfall may alter their survival and growth, leading to a decrease in biodiversity and reduction in mangrove area (Gilman et al., 2008). The issues related to mangroves vulnerability have turned into a serious matter that even scientific community stated mangrove forest may functionally dissapear within 100 years .
Every study represents a valuable contribution to understand the behavior, resilience, and risks related with the future of mangrove ecosystem. Global Land Cover-Mapping studies looks towards the producing scenarios; while regional studies set the basis and provide the tools for local stakeholders in the decision making regarding land-use changes. (Lambin & Geist, 2006). The information about the location of mangroves destruction is essential to determine where mangrove forests reserves are necessaries, and at the same time to comprehend how is the response to environmental and stressor factors in order to set policies of coastal adaptation, resource consumption, or protected areas (Hu, Li, & Xu, 2018 (Gaw, Linkie, & Friess, 2018;Ghosh, Kumar, & Roy, 2017;Godoy, De Andrade Meireles, & De Lacerda, 2018;Mondal, Trzaska, & De Sherbinin, 2018;Nursamsi, 2017;Servino, Gomes, & Bernardino, 2018;and others). In this sense, the purpose of this study is to make use of these remote sensing techniques to determine how mangrove land cover has changed in Parita Bay, Panama.
Panama is situated in the 16 th position of the global ranking of largest mangrove holding nations, with 1,323 km 2 (Hamilton & Casey, 2016). Further, it can be said is the country with the largest mangrove cover in Central America, as per year 1995 (Windevoxhel, Rodríguez, & Lahmann, 1999). However, it has been registered great losses. Actually, the areas of major concern for threatened mangrove species in the world, are found in the Atlantic and Pacific The aim of this study is to monitor the mangrove land cover change in a 32- year period (1987 to 2019) in Parita Bay, located on the Pacific side of Panama.
The main objectives for this study include: • To determine and quantify the mangrove land cover extent change (gain and loss) and its rate of change in the Parita Bay over the last 32 years.
• To determine the changes in quality by using the Normalize Difference Vegetation Index (NDVI) trend of change.
• To identify the possible drivers of change.
• To identify differences in mangrove dynamics between the Protected and Unprotected areas within the study region.
Decision and policy makers need to understand the present, past and future situation of mangrove forests in Panama and the major threats to this ecosystem in order to develop effective regulations plus conservation and restoration projects. Our study looks forward to raise awareness in policy makers regarding the conservation of mangrove forest in Panama.

Study Area
The site of this study is the Parita Bay located between latitude 8°18'40"N and longitude 80°13'37"W to latitude 7°57'18"N and longitude 80°21'49"W in the central Pacific side of Panama ( Figure 1). The area belongs to the called 'Dry Arc' (the driest zone of the country) and is situated within Coclé, Herrera and Los Santos provinces. The study area comprises five (5)  There are two protected areas within the study zone: a. Sarigua National Park. Established in 1984 with an extension of 47 km 2 , although it is referred as of 80 km 2 when including a great part of the sea territory in Parita Bay. More than 50% of the protected area is concessioned to breeding-shrimp ponds for export. Additionally, it is located in an archaeological area of great importance for the study of cultural evolution in the    (Young et al., 2017). However, as these are images from two different sensors,

Data Selection and Image preprocessing
OLI and TM, it should be considered that the difference between them may lead to error in the analysis (Tuholske et al., 2017).
Each band image was clipped in order to extract the free-cloud study region for the analysis. However, for the 2009 period, due to cloud conditions, a composite of two images from December 2009 and January 2010 was created; though some clouds were present, which later were detected and extracted. Each clipped raster band image was used to construct a composite for the classification. The False Color combination of bands Near-Infrared (NIR), RED and GREEN as RGB was used for mapping mangrove on the 1987, 1998 and 2019 images (Islam, Borgqvist, & Kumar, 2018;Jayanthi, Thirumurthy, Nagaraj, Muralidhar, & Ravichandran, 2018).; however, 2009-2010 images seemed to have better results for mapping mangrove on the SWIR, NIR, and RED combination (Gaw et al., 2018;Rahman, Tabassum, & Saba, 2017). Table 1. Data products used for the study

Image Classification
The widely used for mapping mangrove land cover (Heumann, 2011), and is considered as one of the most robust method for classifying mangroves based on traditional satellite remote-sensing data (Kuenzer, Bluemel, Gebhardt, Quoc, & Dech, 2011). The maximum likelihood classifier quantifies the variance and covariance of a spectral class pattern. It computes the statistical probability of a given pixel being a member of a specific land cover class assuming a normal distribution (Lillesand, Kiefer, & Chipman, 2015 the mean, or average, spectral value in each band for each category and then a pixel of unknown identity may be classified by computing the distance between the value of the unknown pixel and each of the category mean (Lillesand et al., 2015). Additionally, it is not the first time for a land cover change analysis to use different algorithms-classified images (Rioja-Nieto et al., 2017).

Field Ground Truth and Accuracy Assessment
We performed a field survey from February 8 th to 12 th 2019 for the accuracy assessment of 2019 image. The reference data must be independent from the data being tested to ensure objectivity of the assessment (Congalton & Green, 2009). For that reason, the field data obtained from the ground truth was not used in the setting of training areas for the supervised classification and was strictly kept for the accuracy assessment. Thirteen places were visited including: 'Monagre' beach, and the protected areas 'Sarigua' National Park and 'Cenegon de Mangle' Wildlife Refuge.
The methodology for the sampling was a stratified random sampling limited to realistic distance from the roads or access ways (Congalton & Green, 2009) because some areas had restricted access or are deep dense mangrove zones, turning dangerous for the team. Using a Garmin 'GPSMAP 64S' we took 330 ground control points (GCP). Afterward, it was built a new set of 125 points inside homogenous class regions nearest to the original GCP, then buffered into a 45 m radius to make sure that the area size is bigger than the small spatial resolution of a Landsat pixel (30m). Additionally, another set of 125 random sampling points was created to complement the previous GCPs from the field, making 250 validation points in total. These points, were later verified using google earth imagery to validate the accuracy of the entire image (Tilahun, 2015). Images from 1987 and 1998 were not validated due to lack of data.

NDVI Analysis
In addition to the land use-cover classification, a Normalized Difference Vegetation Index (NDVI) reclassification was also performed as mangrove cover extent only provides information about area change (quantity). To know about the quality change (greenness), NDVI time series map is a good resource (Alatorre et al., 2015). The NDVI is the most widely used and robust index for vegetation analysis (Kuenzer et al., 2011;Wulder & Franklin, 2003). This index provides information about the photosynthetic capacity of absorption of plants and leaf resistance to water vapor transfer (Ruimy, Saugier, & Dedieu, 1994 This information is applicable for the NDVI of all classes in the study area and may include some bias mainly due to crop cover, between harvested and nonharvested fields. However, to know the changes in NDVI of the mangrove cover specifically, it was performed a random sampling of 500 points in the 'mangrove' layer with NDVI values (Figure 3). Finally, these points were displayed in boxplots, to know the trend of change of NDVI in Mangrove class.
Same procedure was performed for 'Other vegetation' cover in Parita Bay.
. Figure 3. Random sampling of 500 pixel points of NDVI values in unchanged mangrove cover.

Land Cover-Use Change (LUCC) detection
Change Detection for the classified Land Use-Cover images and the other reclassified NDVI maps was developed by using the Semi-Automatic plugin (Luca Congedo, 2016) and MOLUSCE (Modules of Land Use Change Evaluation) plugin (Rahman et al., 2017) in QGIS 3.4 and 2.18.

Analysis in Protected Areas
Additionally, a Land Use-Cover Change detection for Protected and

Analysis of Environmental variables in the study area
To investigate the possible drivers of change of the mangrove cover in Parita Bay some climatic variables were analyzed. Rainfall, Mean, Maximum and

Classification Accuracy
The overall accuracy for the 2019 classified images is 87% with a Kappa coefficient 0.83 which is considered a good level of agreement between classifiers (Wulder & Franklin, 2003) (Table 2). Classification for the 2009-2010 image also shows a strong agreement between the reference and classified data as Overall Accuracy is 90% and Kappa coefficient 0.86 (Table 3). In general, it can be said that most errors in the mangrove cover classification fall upon the Producer (error of omission) rather than the User (errors of commission); in other words, mangrove class tends to be underestimated in the map. For example, one source of error we found on field is that some mangroves have a small size and are regenerating in the middle of a very dry bare soil near 'Sarigua' and 'Cenegon de Mangle' Protected Areas. Therefore, in the map, it is classified as bare soil class, but in the field, it seems as mangroves regenerating, consequently, it is identified as mangrove class, generating errors of omission. For both images '  tends to have the lowest accuracy, mainly due to the crops, which appear like bare soil class in the classifier because have been harvested, while in the reference may appear as crops. However, 'Bare soil & Built-up' is not a category of concern for this study as this analysis is more focus in Mangrove and Aquaculture & Salt-Pan (AS) classes.

Land Use-Cover Change (LUCC) Detection and mangrove estimation in Parita Bay
Land Use-Cover classified maps are shown in Figure 6. Over the 32-year analysis, mangroves had increased in extent more than 500 hectares (4.7%), with an annual average rate of change of 0.15%. However, during the period from 1987 to 1998 there is an evident decrease in mangrove cover (151.83 ha) with a rate of -0.11% (Table 4). In spite of this decrease, eventually mangroves

NDVI Classes Changes on time
Similar to the Land Use-Cover change detection, NDVI classes' time series analysis was also developed (Table 7 and Figure 9). This will help us to understand the changes of NDVI along the entire study area. One thing to consider is that this information may include some bias mainly due to crop cover,

NDVI trend in mangrove and other vegetation covers
The sampled NDVI values for the mangrove cover present a skewed left distribution, notice on the boxplots (Figure 11), which tells us that mangroves on the site present high or very high NDVI values (close to 1).      (Tables 8 and 9).

Mangroves changes in Protected vs. Unprotected Area
Matrix of change in mangrove cover for each protected area reveal that the major class producing mangrove losses and gains is 'Other vegetation' (Tables 10 and 11).

Rainfall
The One-way ANOVA test between stations show there is not significant difference between Parita and Los Santos station (p= 0.5) and between 'Puerto Posada' and 'Rio Hondo' stations (p=0.6), but there is a significant

Temperature
Same as rainfall data, Average Annual Maximum, Mean and Minimum local Temperature was calculated ( Figure 18)

IV. Discussion
While generally other studies have reported a mangrove net loss in their and Parita Bay also struggle through it 1 . Rainfall, in particular, has an effect on the salinity of mangrove soils, especially at low tides and high evaporation rates, because rainfall dilutes and leach salt. While in arid conditions (low precipitation), the salt tends to concentrate more (Lüttge, 2008 can also produce changes in the mangrove cover behavior (Day, Allen, Brenner, Goodin, & Faber-langendoen, 2015;Valiela, Elmstrom, Lloret, Stone, & Camilli, 2018). However, the analysis of these other variables are out of the scope of our work.
On the other hand, during the same period (1987)(1988)(1989)(1990)(1991)(1992)(1993)(1994)(1995)(1996)(1997)(1998) Classification results for the next two decades (1998-2019) reveal an increase in mangrove cover of almost 6%. It is important to highlight the stability and persistence of mangroves in Parita Bay, as they recovered from a loss period by increasing in area and greenness; considering there are not strong and updated regulations, coupled with the impact of bioclimatic factors, as this is the driest zone in the Country. In general, mangrove forest have strong resilience characteristics (Alongi, 2008).
Over decades and centuries, mangroves have been considered not to conform to typical ecological patterns of succession (Lovelock, Sorrell, Hancock, Hua, & Swales, 2010). In our study, visual inspection of remote sensing imageries reveals seaward expansion or seaward colonization in some points of the coast of Parita Bay. Mangroves accumulate great amount of sediment through peat formation rather than just for stabilization, enough to call them vertical land-builders (Lee et al., 2014).
These facts and arguments might let us assume that mangroves in Parita Bay possess a strong resilience. However, it must not be discarded the fact that this increase in area and good health status (High NDVI) might be also affected by reforestation, and restoration activities in the zone.
Reforestation-Afforestation programs are weighty factors involve in mangrove regrowth (Jayanthi, Thirumurthy, Nagaraj, Muralidhar, & Ravichandran, 2018;Nursamsi & Komala, 2017 Although it remains unclear the role of the management system of these two protected areas in this matter, it recalls the need to evaluate them, and an appropriate tool recommended is the Management Effective Evaluation 3 . In addition, the Sarigua National Park is exposed to high wind erosion regimes (Cooke & Ranere, 1992), and has been pointed as facing one of the most severe process of soil degradation in the country (ANAM, 2000), together with elevated temperature, hypersalinity, and low precipitation regimes (MiAmbiente, 2014). In other words, this is one of the most critical environments in the Country and it needs special attention.
The comparison test for rainfall data shows that meteorological stations closer to the protected areas registered less precipitation regime than those farther from the protected sites. However, to relate precipitation and temperature with the information based on satellite imageries (for example, NDVI), it is necessary a major number of scenes. The problem relies in that getting a major number of free cloud imageries in the tropic regions is a very difficult or maybe impossible task for low temporal resolution satellites like Landsat.

IV. Conclusion
This study represents the first time change monitoring of mangrove forest cover in the country after the year 2000; and is the first time-series analysis of mangrove forest cover in Parita Bay. In resume, mangroves in the Parita Bay region have increased in extent over the last 32 years. However, during 1987 to 1998, mangroves experienced a net decrease in area and greenness, which might be influenced in part by natural factors, such as rainfall and temperature. Also possibly linked to anthropogenic activities such as aquaculture, saltpans, as well as agriculture and other change practices. Nevertheless, there might be other drivers of changes such as pollution from upstream, tidal range and sociodemographic characteristics, which are not included in our analysis. The established policies and regulations for mangrove conservation and protection may have not influenced directly into the overall history of change, but if more strong and updated regulations are implemented, the mangrove cover increase rate may be helped. Moreover, unidentified reforestation programs may be hidden drivers of change, and it is necessary to know where these areas are located through an open platform, because we can misattribute a mangrove regrowth to its own resilience when it is due to an enrichment activity in the zone.
Turning to the issue about protected areas, mangrove cover has decreased during the last decade in Cenegon de Mangle Wildlife Refuge and Sarigua National Park while in unprotected regions it increased, which is a matter of concern. In addition, further research is suggested along with a profound comparison study of Land Use-Cover Change in places holding mangrove forests under the status of protection versus non-protected parts in the country. This will contribute to determine a generalizable overview of protected mangroves in Panama. Finally, we hope this study would become a baseline for policy makers to protect mangrove ecosystems in the Pacific Coast of Panama and set goals for conservation and monitoring of mangrove reforestation and restoration programs.