Isabela Land Cover Assessment Process Documentation

Isabela Land Cover Assessment Process Documentation

Environmental Science for Social Change

December 2020

Isabela Land Cover Assessment Process Documentation 1

Table of Contents

Introduction.............................................................................................................................3

Project Objectives...................................................................................................................3

Data and Mapping Method....................................................................................................3

The Sen2Agri System........................................................................................................3

Workflow for 2020 Isabela Land Cover..............................................................................4

Data Downloading and Preprocessing...............................................................................4

In Situ Data Generation.....................................................................................................5

Classification Process............................................................................................................6

Crop Type Map Generation (with in-situ data)....................................................................6

Post Processing.................................................................................................................7

Validation: Online Mapathon and GIS Training with Stakeholders..........................................7

References.............................................................................................................................8

Annexes...............................................................................................................................10


IntroductionIsabela is the second largest province in terms of land area, and one of the last places with contiguous low altitude tropical rainforests in the Philippines. It also comprises the NorthernSierra Madre Natural Park (NSMNP), the largest terrestrial protected area in the country. As one of the last forest bastions of greater Luzon bioregion, Isabela has a huge role to play in ensuring an integral ecology of this landscape, yet it is one of the most threatened as well.

A comprehensive baseline assessment of the land cover does not exist for the Northern Sierra Madre landscape. Although available geospatial data from the government exists, these do not have the variety of land cover/land use types that would give a more accurate view of the newly emerging threats to land and forest cover change.

As part of ESSC’s continuing effort on forest and land cover update, the Isabela Land Cover Assessment (ILCA) was launched in June 2019 with the support of Forest Founda-tion Philippines (ESSC, 2019). Its main goal is to contribute to forest monitoring and land cover change in the Sierra Madre by providing robust baseline using field data and remote sensing.

Project ObjectivesAmong the objectives of ILCA are to undertake;

1. Data acquisition, preparation, and input for around 1.2 million hectares2. Analysis, validation, and finalization3. Knowledge generation and method documentation4. Knowledge sharing and collaboration

Data and Mapping Method

The Sen2Agri SystemThe Sentinel 2 for Agriculture (Sen2Agri) is an operational standalone system that uses Sentinel 2 (A&B) and Landsat 8 images to produce remote sensing products for agriculture (Defourny et al., 2019; European Space Agency [ESA], 2019; Inglada et al., 2015; Matton et al., 2015). Though used mainly for agriculture, the random forest algorithm implemented for generating Crop Type Map (L4B) product can be tweaked to include all types of land cover classes such as forests, built-up areas, and others. The system’s automatic mode fordownloading and processing satellite data enables the user to take advantage of the high temporal (every five days) and high spatial resolution (10 meters) of Sentinel 2 satellites with less effort. Also, the system is an open source and freely available through the GNU General Public License.


Workflow for 2020 Isabela Land CoverThe figure below shows the general workflow in generating the 2020 Isabela Land Cover Map. First, the first process involves the downloading and pre-processing done by Sen2A-gri system. This process had minimal user interface, and was automatically done by the system. Before going to the classification processing proper, in-situ data were generated through windshield/motorcycle surveys and photo interpretation. The in-situ data will be used as an input for the land cover processing step. After that, any preliminary output mapswill be validated on line and through virtual consultation meetings/mapathons with local stakeholders. Any insights that will improve the map outputs were then integrated in the ex-isting in-situ data. In this way, throughout the mapping process the outputs are enhanced until in reaches the acceptable level for all users.

Fig. 1 General Workflow for 2020 Isabela Land Cover.

Data Downloading and PreprocessingThe downloading and preprocessing phase of Sen2Agri requires minimal user intervention. With each new acquisition of available imagery, the system downloads them and performs preprocessing step such as atmospheric and terrain correction. For Isabela, there are four tiles covered namely, 51QUV, 51PYK, 51NXJ and 51NYJ (Figure 1). In the 2018 up to 2020 season, there were a total of 589 downloaded images but only 350 images were suc-cessfully processed by the atmospheric correction. The remaining 239 were excluded sincethese images have more than 90% cloud cover.


Fig. 2 Sentinel 2 tiles covering Isabela province.

Table 1. Number of Sentinel Images used for the Classification

Number of Images Description

350 Successfully processed by atmospheric correction.

239 Invalid Images (High Cloud Cover)

589 Total Images Downloaded

In Situ Data GenerationIn situ data is similar to the terms such as “training data”, “vector polygon” or “training poly-gon”. The goal of in situ data generation is to have an updated and robust input data that reflects the current land cover situation from the ground. This input data will train Sen2A-gri’s random forest classification algorithm. It will also serve as validation polygon that is re-quired to determine the accuracy of the output map.

The general rule in creating in situ data is to have a general knowledge about the size, pro-portion and distribution of land cover of the province of Isabela. Based from the 2015 Na-tional Mapping and Resources Information Authority [NAMRIA] land cover data, about 40%


of the province is classified as forests (portion of the Sierra Madre Mountain Range). The rest of the classes are distributed to annual, perennial and other land cover. Major land cover classes (corn, rice, forests ) need to have a minimum of 100 feature and up while mi-nor classes (built-up, shrubland, grassland) are required to have a minimum of 50 features in order for them to appear in the output map.

A total of 13,611 features were generated to form the in situ training data. It was derived from 13,088 geotagged photos captured from windshield/roadside collection surveys (Waldner et al., 2019). It is important to take note that the geotagged photos were also veri-fied using updated high resolution images from Google Earth and Google StreetView.

Fig. 3 Sample screenshot of in-situ data in Isabela.

Classification Process

Crop Type Map Generation (with in-situ data)Through the web interface of Sen2Agri, the user will either submit custom jobs or just up-load the in-situ data in the sites section and wait for automatic creation of L4B (Crop Type Map) products (Udroi et al., 2019). In the automatic mode, users will have two products generated over a single season. It is recommended to use the custom jobs mode to focus in processing just one output given a single training data. Depending on the season dura-tion and computer specs, a minimum of 12-24 hours are needed for the processing to fin-ish.

Post ProcessingThe post processing steps include merging, sieving and clipping. The raw raster ouputs of Isabela 2020 Land Cover maps include four tiles that requires merging. Next, to lessen the salt and pepper effects of the output, sieving is applied to the merged tiles with a threshold


of 40 pixels. Finally, the raster outputs were clipped using the official provincial boundary vectors coming from the Provincial Government of Isabela.

Additional notes on the inclusion of mossy/montane forests:In mapping the mossy/montane forest, an elevation cutoff from Digital Elevation Models (DEM) of 900 meters above sea level was chosen for Isabela province. The said elevation cutoff was applied to the output land cover map where the forest types should be primary forest, thus excluding the secondary forests and other forest formation (ultramafic and lime-stone forest types). The 900 masl cutoff was used to reflect the mass elevation effect of mountains near the coast and deemed as a midway between the 1,000 and 800 masl ele-vation where most of mossy and montane forest where found (Mabuwaya Foundation, 2018). This was done to take into account the Massenerhebung effect or mass elevation ef-fect (Grubb, 1971).

Validation: Online Mapathon and GIS Training with StakeholdersThere were two online strategies used to compensate for the face to face validation and training of Isabela Province Partner Organizations. The first is a series of Online Mapathon where local government units (LGUs) can freely and actively participate in suggesting com-ments and recommendations to improve the accuracy of map features based from their ground and field experience. Development in the recent years were also discussed to iden-tify the probable land cover change drivers on each municipalities/cities.

On the other hand, Online GIS Training for the Isabela partner organizations primarily served as a venue to solicit enhancements and additional land cover features for the train-ing data. Basic GIS training were also carried in the form of blended learning (webinars, self paced exercises) so that participants can have basic GIS skills which were eventually applied in improving and adding features on the training data.

July 20, 2020: 1st run after adding 990 features in San Mariano, llagan City and Cauayan City (c/o Mabuwaya Foundation: Bernard Tarun and Mario Pedrablanca, Provincial Government of Isabela: B. Pintucan, Cagayan Valley Research Center: Eugine Ramos) from 1st Online GIS Training and Training Data Enhancement An-nuals Duration: April to December 2020, Perennial: April 2018-February2020, total features (12,511)

Aug 12, 2020: 2nd run after adding 481 features from second GIS training contrib-uted mostly by LGU Jones with 387. Other LGUs who contributed: Angadanan, Cauayan City and San Mariano, with additional from Cagayan Valley Program for Environment and Development (Ryan dela Cruz)

October 28, 2020: 3rd and final run after the Online Mapathon Discussions with LGUs (LGU list available in quarterly report ) has resulted to a total features of 13,611 training polygons.


ReferencesDefourny, P., Bontemps, S., Bellemans, N., Cara, C., Dedieu, G., Guzzonato, E., Hagolle,O.,

Inglada, J., Nicola, L., Rabaute, T., Savinaud, M., Udroiu, C., Valero, S., Bégué, A.,Dejoux, J.-F., El Harti, A., Ezzahar, J., Kussul, N., Labbassi, K., ... Koetz, B., (2019). Near real-time agriculture monitoring at national scale at parcel resolution: performance assessment of the Sen2-Agri automated system invarious cropping systems around the world. Remote Sensing of Environment. 221, 551–568. https://

doi.org/ 10.1016 j.rse.2018.11.007

Environmental Science for Social Change (2019). Isabela Land Cover Assessment (ILCA). Retrieved from https://essc.org.ph/content/archives/11150/

European Space Agency (2019). Sentinel-2 for Agriculture. Retrieved from http://www.esa- sen2agri.org/ .

Grubb, P. J. (1971). "Interpretation of the 'Massenerhebung' Effect on Tropical Mountains". Nature. 229 (5279): 44–45

Inglada J., Arias M., Tardy B., Hagolle O., Valero S., Morin D., Dedieu G., Sepulcre G., Bontemps S., Defourny P., & Koetz B. (2015). Assessment of an Operational Systemfor Crop Type Map Production Using High Temporal and Spatial Resolution Satellite Optical Imagery. Remote Sensing. 7(9):12356-12379. https://doi.org/10.3390/

rs70912356

Mabuwaya Foundation (2018). The Northern Sierra Madre Natural Park. Quezon City Philippines.

Matton N., Canto G. S., Waldner F., Valero S., Morin D., Inglada J., Arias M., Bontemps S.,Koetz B., & Defourny P. (2015. )An Automated Method for Annual Cropland Mappingalong the Season for Various Globally-Distributed Agrosystems Using High Spatial and Temporal Resolution Time Series. Remote Sensing. 7(10):13208-13232. https://

doi.org/ 10.3390/rs71013208

National Mapping and Resources Information Authority (2015). Land Cover 2015 of IsabelaProvince [Shape File]. Fort Bonifacio, Taguig City Philip pines

Udroiu, C. Nicola, L, Cara, C., Savinaud, M., Bellemans, N., Malcrops, P., … Defourny, P. (2019) Sentinel 2 for Agriculture Software User Manual ver sion 3 [PDF File]. Retrieved from http://www.esa-sen2agri.org/wp-con tent/uploads/ Sen2Agri-Soft

ware-User-Manual-3.0.pdf

Waldner, F., Bellemans, N., Hochman, Z., Terence, N., Abelleyra, D. Verón, S.R., … Defourny, P. (2019). Roadside collection of training data for cropland mapping is viable when environmental and management gradients are surveyed. International Journal


https://doi.org/10.3390/

https://doi.org/10.1016

https://doi.org/

https://doi.org/

http://www.esa-sen2agri.org/wp-content/uploads/

http://www.esa-sen2agri.org/wp-con

https://doi.org/

https://doi.org/

of Applied Earth Observation and Geoinformation, 80, 82-93. https://doi.org/10.1016/ j.jag.2019.01.002


https://doi.org/10.1016/

AnnexesAnnex 1. Isabela 2020 Land Cover


Documents

Isabela Land Cover Assessment Process Documentation