SPATIO-TEMPORAL WATER QUALITY MAPPING USING SATELLITE DATA AROUND A MANGROVE PLANTATION IN CAGSAO, CALABANGA, CAMARINES SUR

The Cagsao mangrove is a thriving young forest along the San Miguel Bay (SMB), Camarines Sur. To establish the Spatio-temporal Water Quality mapping, data from the Chesapeake Bay, an estuary in the United States of America (USA), was sourced as the train set for this study. Spatio-temporal maps of chlorophyll and dissolved oxygen were generated using Linear Regression (LR) models which were derived from the train set and satellite images of the SMB. GNU (GNU’s not UNIX) Octave was used for the image processing, computing, and analysis. There were three phases in the image processing conducted in this study, 1) extraction of image data of the corresponding measure points from the train area, 2) conversion of the satellite study area to a two-color raster image, and 3) generation of the spatio-temporal maps from the analysis. The study found that the SMB is in the range of Mesotrophic to Moderate Eutrophic classification. The decay from two other point sources (Manga River and Libmanan River) was compared to that of Tigman River, an adjacent river to the Cagsao mangrove forest to determine variations and impact of the mangrove forest in the water quality of the SMB. The presence of Cagsao mangrove forest was found to affect the gap of increasing chlorophyll levels from shore toward the bay center in the adjacent Tigman River unlike Manga River and Libmanan River, which have both no adjacent mangrove forest in the river mouth area. The corresponding satellite images for the dataset taken during and near the date of the train area measurements were also extracted.


Coastal Waters and the Mangroves
The annual impact of economic and social activities in Coastal Regions amounting to billions of dollars in the global scene (Costanza, et al., 1997;Martinez, et al., 2007) shows the importance and need for sustainable coastal water management. It was estimated that several kilometers per year of aquaculture areas can erode posing danger to coastal communities (van Wesenbeeck, et al., 2015).
Apart from potential benefits to ecotourism, these ecosystems also provide researchers with various habitats for the conduct of extensive coastal research (Mallin & McIver, 2012).
One of the natural defenses available to ocean degradation is the presence of mangrove forests. A mangrove forest commonly serves as a habitat for juvenile aquatic animals and other marine animals which seek refugee during periodic seasons or typhoons. The most recognized ecosystem services of mangrove forests include serving as a carbon sink and as ocean greenbelts protecting coastlines from from tsunamis and storm surges (Alongi, 2008;Barbier, S. D. Hacker, Koch, & Stier, 2011). These forests are critical agents for coastal area protection (Veettil & Quan, 2019) and curb climate change effects. However, over 35% of the total mangrove cover worldwide, has disappeared due to clearcutting and aquaculture ponds conversion since the 1980s (Richards & Friess, 2016).
One of the important indicators of a thriving marine life is the level of plant-producing organisms and dissolved oxygen present, as a source of food and oxygen, respectively. Phytoplankton are microscopic marine algae which hold the base of several aquatic food webs. Also called microalgae, they contain chlorophyll and need sunlight, hence they live in the surface waters (NOAA, 2019). Marine productivity and capture of greenhouse gases (GHGs) are closed-related to the function and structure of planktonic communities (Karl, Bidigare, & Letelier, 2001).
The regulation and variability of the concentration of dissolved oxygen are primary to the function of the marine ecosystem since it supports marine life. It was found in the correlation maps of the Mediterranean Sea that the DO in the surface waters were dependent on solubility with some temperature factor in contrast to the deep waters where DO was primarily influenced by dynamical processes like density, circulation, or transients (Mavropoulou, Vervatis, & Sofianos, 2020).

Water Quality Monitoring
Biomonitoring is essential in cases of water pollution accidents (Schalie, Shedd, & Widder, 2001). Principal component analysis is generally used to interpret the characteristics of water. It works by separating data points into clusters while establishing relationships among the different variables (Qu & Kelderman, 2001). Several water quality studies focus on temporal analysis (Astel, Mazerski, Polkowska, & Namieśnik, 2004) and spatial variation mapping (Astel & Simeonov, 2011). The use of septic weight index was proved as an accurate predictor in near shore water quality as well (Norat-Ramírez, P.Méndez-Lázaro, E.A. Hernández-Delgado, H.Mattei-Torres, & L.Cordero-Rivera, 2019). Numerical modeling was found to be an important tool in the planning, management, and decision-making to reduce wastewater discharge (Doan, Nguyen, Quach, Tran, & C.D.Nguyen, 2019). In data-driven modeling, the most pressing issue is the creation of a probabilistic model with logical arguments and its accuracy to represent a phenomenon (Kim & Park, 2017).
Spatio-temporal mapping is the key to understanding the variations in water quality parameters and maintaining the health and safety of both the water and aquatic ecosystems (Chu, Kong, & Chang, 2018).
However, in-situ monitoring is labor-intensive and is not viable in some areas, especially during spawning months. The combination of using Remotely Sensed (RS) data with in-situ water quality measurements has provided both a significant and cost-effective evaluation of water quality variations (Devlin, et al., 2015) and monitoring.

The Mangrove Plantation in Cagsao
With approximately over 6 hectares of mangrove cover, the study area is a once deserted shore part of San Miguel Bay fronting the Pacific ocean, in the province of Camarines Sur (Faustino, Madela, Castor, Muroda, & Chavez, 2020). Specifically, the mangrove forest is in Barangay Cagsao, Calabanga, a coastal village where several houses were lost during a storm surge back in 1998 (Soltes & Rodriguez, 2010). It was the year 2009 when volunteers started coming to plant mangroves in the area including students, teachers, health professionals, and even foreigners. Various groups found interest in the rehabilitation of the mangrove and in cooperation with the LGU of Calabanga and DENR, Smart Communications spearheaded the rehabilitation. It was as early as 2010 that increases in stocks of crablets were reported, then a ban on the gathering of stocks inside the plantation was banned (Escandor, 2012). The village fishermen are convinced that the mangrove forest provides for the community, it allowed them to catch more fish (Soltes & Rodriguez, 2010).  (Nasa, 2020) Local institutions like the Central Bicol State of Agriculture (CBSUA) and Ateneo de Naga University (ADNU) are regularly conducting mangrove planting activities in the area and other mangrove sites in Calabanga to date (Faustino, Madela, Castor, Muroda, & Chavez, 2020). Shown in Figure 1 the Satellite image of SMB.

Train Area
Home to approximately 3600 species of plants and animals, Chesapeake Bay is the largest estuary in the United States. This bay stretches across six states and is being regularly monitored due to its economic and biodiversity importance (Chesapeake Bay Foundation, 2021).  (Nasa, 2020) The main goal of the study is to use satellite data for a Spatiotemporal water quality mapping around the Mangrove Plantation in Cagsao, Calabanga, Camarines Sur to determine the variations in the water quality surrounding a mangrove forest using Train data from the Chesapeake Bay. As shown in Figure 2, the Satellite image of the Chesapeake Bay with the plotted measure points.

Image Processing
To maximize the temporal property of the satellite images, cloud-free images with acquisition dates that are close to the sampling dates of the train area water quality (TAWQ) data were preferred. Masks were made to separate water from land or clouds. Focal mean filtering was also used to enhance spatial patterns and generate a noise reduction. Grey-scale images of all bands were examined first visually to provide an added layer of quality control. It was found that visible bands contain the most information about the spatial distribution of surface waters as reflected in the satellite imagery. Principal component analyses (PCA) were performed to concentrate the important characteristics into one layer and facilitate better recognition of spatial patterns. (Erkkila & Kalliola, 2004).
GNU Octave was used in the image processing for the study. It is an open-source alternative to Matlab. Image and IO packages were incorporated to perform image processing and extraction of data from MS Excel. Satellite images from NASA with Terra, MODIS, True Color, Corrected Reflectance base layer were used. Cloud-free images were preferred to avoid further image manipulation. Images with excessive cloud cover were replaced with cloud-minimum images taken nearest to the measure dates. There are two sets of satellite images used, the train area (Chesapeake Bay) and the study area (San Miguel Bay).  The first phase of image processing was the extraction of image data of corresponding measure points from the train area CB as shown in Figure 3. These data were needed as inputs in the Linear Regression (LR) Model. All three reflectance, Red, Blue, and Green channels, using 8bits each with values 0 to 255 was extracted. There are a total of 6 satellite images (approximately 70% of the train images) used in this phase.
The second phase of image processing was the conversion of the study area SMB (Terra, MODIS,7-2-1, Corrected Reflectance) satellite imagery to a two-color raster image separating the water (R-255, G-0, B-0) from land (R-255, B-255, G-255). By separating land from water, the coordinates of the study area needed for mapping were extracted. Phase 2 is illustrated in Figure 4.

Figure 5. San Miguel Bay Study Images
The third phase uses 12 satellite images of the study area spanning for 3 years, 2017-2019, taken quarterly as shown in Figure 5. From the coordinates generated in the second phase, the image data of the bay for all 12 images was extracted. Only the water data was extracted and cleaned for the next phase. The last phase of image processing was the generation of the map reflecting the Spatio-temporal WQ model for chlorophyll and DO for all 12 images. The model-derived values of Chlorophyll and DO were mapped following a color gradient intensity from blue to red.

LR Model
Usage of the red, green, and blue reflectance as inputs yields better model results (Chu, Kong, & Chang, 2018). Linear regression was employed to determine the pattern of water quality around the perimeter of the mangrove reserve. Data extracted from image processing phase 1 were used as variables R, G, and B respectively.
Y value from general equation 1 is the Chlorophyll and DO values, respectively. There are two models generated, the CH and DO models. Though other studies found that the Geographically Weighted Regression (GWR) models yield better results than LR (Chu, Kong, & Chang, 2018), establishing the LR models is equally important, since this practice is common (Hellweger, P. Schlosser, & Weissel, 2004), specifically for lakes and reservoirs (Kloiber, Brezonik, & Bauer, 2010).   Table 1, the LR Model coefficients from the Linear Regression analysis, with 32 observations or data points, from the train set CB. Errors were not incorporated in the LR model. It can be noted that though the B band appears to be statistically insignificant due to its p-value, it was still included in the model due to the uncertainty of the dataset variation should an in-situ measurement for the study area be available. Shown in Figure 6-7, the residuals and fit plots for Chlorophyll and Dissolved oxygen, respectively. Both residual plots of chlorophyll and dissolved oxygen in Figure 6, the random dispersal is present only in the range of 0 to 100 for R, G, and B bands. Normally, when randomness in the dispersal is not observed, a non-linear regression is appropriate. However, this may not be the case for this data since there is a limitation in the trains set data used. As shown in Figure 7, the fit plot of dissolved oxygen is better predicted than the chlorophyll. This gap can be accounted by the uncertainties on other factors affecting phytoplankton migration like illumination (Gerbersdorf & Schubert, 2011).

Shore Decay Points
From Table 1, the LR models were represented by Equations 2 and 3. Both equations were plugged into GNU Octave where computing and image processing and maps were generated.   The decay shore decay points as illustrated in Figure 8 start with the three (3) rivers, Manga, Libmanan, and Tigman. The MR1, LR1, and TR1 are points at the river mouth while MR3, LR3, and TR3 are points near the bay center.
A three-year data was generated from the quarterly extracted satellite imagery, as processed, as reflected in Table 3-4.  The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XLVI-4/W6-2021Philippine Geomatics Symposium 2021, 17-19 November 2021 Shown in Figure 9-10 the illustration derived from LR models of Chlorophyll and Dissolved Oxygen for the decay points illustrated in Figure 8. It can be noted that the Chlorophyll shore decay for Tigman River presented a larger gap between TR1 (river mouth) and TR2 while substantially the same levels in TR2 and TR3. A different observation was present to Manga and Libmanan River which have no adjacent mangrove and have a mangrove forest along banks, respectively. Figure 11. Quarterly Spatio-temporal Map of SMB -Chlorophyll (2017-2019 Illustrated in Figure 11 the Spatio-temporal maps of a portion of SMB subjected to model-derived chlorophyll and dissolved oxygen data. It can be observed that the streamflow of Libmanan River highly affects the level of Chlorophyll present in the waters surrounding its mouth. While the Tigman River, which is smaller in size, appears to have a smaller range of influence on the chlorophyll levels of the surrounding waters. This can be attributed to the presence of the Mangrove Forest acting as a natural wastewater treatment relative to the adjacent Tigman River. The Manga River, however, does not show a significant effect on the chlorophyll levels of the surrounding water which can be attributed to the wind direction during this time of the year (MeteologixPH, 2020). In Figure 12, the Spatio-temporal DO map shows an inverse of the Chlorophyll map. As shown in Figure 13, the side-by-side Chlorophyll and DO Spatio-temporal Map of SMB shows their inverse relationship.

CONCLUSION
Though LR models were only employed, the generated Spatiotemporal map showed a better illustration of the variations of water quality in San Miguel Bay both for chlorophyll and dissolved oxygen than a one-point comparison. It was clearly presented that the dissolved oxygen is inversely proportional to the chlorophyll present in the surface waters, yet both are important factors in the productivity of aquatic ecosystems such as in San Miguel Bay.
Analysis of shore decay has presented an overview of how DO disperses from the river mouths toward the bay center. This map can also indicate activities that happen in the bay and reflect phenomena and seasonal variations that may affect spawning and fish catch.
Further, the chlorophyll levels in Figure 12 indicate a Mesotrophic to Moderately Eutrophic lake classification. But it is apparent that the Cagsao Mangrove forest played a part in increasing the gap of chlorophyll level variations in Tigman river mouth toward the bay center, unlike Manga and Libmanan river which have no adjacent mangrove and have a mangrove forest along banks, respectively. The levels of dissolved oxygen on the other hand are acceptable with DENR standards for class AA, highest usage classification (DENR, 2016).
To improve the study, the use of more observations and conduct of in-situ water quality monitoring is highly recommended. Measurement of other water quality parameters is also encouraged to be able to deepen the understanding of the variations and patterns of decay in an aquatic ecosystem.
Rechecking of the residual is also recommended to determine whether the linear regression will be appropriate given a larger dataset. In the future, the placement of monitoring systems with real-time data drop will serve as an avenue to advance remote sensing research on water. Intensive shore decay analysis is highly recommended in comparison with historical phenomena such as spawning, low fish catch, fish kill, and increase in river mouth nutrients. Finally, uncertainties like the wind effect must be considered in the dispersal of water quality parameters and dynamics of aquatic organisms, specifically the plankton.

ACKNOWLEDGEMENTS
I would like to acknowledge Dr. Yap and Dr. Mari for inspiring me to get a grip and write this paper amid the COVID-19 pandemic.