Technical Documentation

Ensuring ecological stability and biological productivity over large areas is critical for landscape management and conservation policy, USAID/Cambodia, along with other stakeholders, require insight and reports on the effectiveness of biodiversity and conservation-related project interventions. SERVIR SEA’s Biophysical M&E Dashboard tool was developed in response to those needs. The Cambodia Biophysical M&E Dashboard is the tool that can monitor performance of landscape-scale efforts and biophysical conditions on-the-ground. SERVIR-Mekong developed the Biophysical M&E tool to meet the request from the USAID mission in Cambodia. Development of the tool designed for project monitoring and evaluation, enabling the identification of changes in biophysical conditions on-the-ground to support USAID-funded biodiversity and conservation projects.

This manual provides information about the framework and approach that has been applied to build the Cambodia Biophysical Monitoring and Evaluation Dashboard, and the structure of components in the tool. The tool leverages the power of cloud computing and open satellite data on Google Earth Engine for data processing and monitoring the change of biophysical condition and forest cover in Cambodia due to conservation project intervention and support for sustainable landscape management. This User Manual therefore, contains all essential information for the user to make full use of the Cambodia Biophysical M&E Dashboard tool by providing stepwise information on visualizing and exploring the platform and step-by-step procedures for information access and use.

Introduction to Cambodia Biophysical M&E Dashboard

Cambodia is home to around 50 protected areas and a number of biodiversity corridors, which cover an area of approximately 7.5 million hectares - equating to 41% of the country’s total land area. Cambodia’s protected area system contributes positively to the country’s economy and sustainable development through the sustainable use of its biological, natural, and cultural resources. Through implementation of USAID’s Supporting Forests and Biodiversity (USAID SFB) projects in Cambodia, there have been numerous meaningful results that have improved the conservation and governance of Cambodia’s forest landscapes.

Figure 1.Launching page of the Biophysical Monitoring and Evaluation Dashboard

SERVIR SEA developed the Biophysical M&E tool to meet the request from the USAID mission in Cambodia. Development of the tool designed for project monitoring and evaluation, enabling the identification of changes in biophysical conditions on-the-ground to support USAID-funded biodiversity and conservation projects, according to the USAID indicators outlined below:

Indicator EG.10.2-1: Number of hectares of biologically significant areas showing improved biophysical conditions as a result of USG assistance. Area of improved or degraded biophysical condition in the project area.

Indicator EG.10.2-2: Number of hectares of biologically significant areas under improved natural resource management as a result of USG assistance. Annual area of deforestation (ha) and/or reforestation (ha).

Therefore, the Biophysical M&E Dashboard tool builds upon SERVIR-Mekong’s existing Eco-Dash platform and consists of four key components. The default ‘biophysical monitoring’ layer evaluates changes in broad-scale biological productivity using the Enhanced Vegetation Index (EVI), a measure of relative biomass particularly suited to high-biomass areas of the globe, derived from MODIS multispectral data. The other three components are Forest monitoring (forest gain/loss, derived from Landsat); Forest alerts (threat alerts, derived from Landsat); and Fire hot spot monitoring (hotspots and burned area, derived from MODIS). The user interface leverages the power of Google Earth Engine (GEE), a cloud computing platform that links extensive data archives with substantial processing power. With the continuous time-series feature, the user is able to select a baseline and evaluation period for each of the components as well as an identified area of interest which allows for easy comparisons between ‘before’ and ‘after’ intervention timeframes (Poortinga, 2018). The results can be exported in report-style format enabling decision-makers and non-technical users alike to directly use this analysis and information. These components of the tool were selected based on the needs of USAID/Cambodia and their specific M&E indicators, with carbon and biomass estimations to be incorporated in the future.

Application Purpose

Aside from monitoring performance of landscape-scale efforts and biophysical conditions on-the-ground, the Landscape Monitoring Dashboard Tool can provide valuable insights into the possible drivers of change such as climate change, urban expansion and infrastructure development.

Application Uses

The Landscape Monitoring Dashboard tool is currently being used by USAID/Cambodia to monitoring performance of landscape-scale efforts and biophysical conditions on-the-ground, monitor the change of the landscape and rice cultivation on-the-ground for food security purpose in Cambodia.

Tool Approaches and framework

The Cambodia Biophysical Monitoring and Evaluation Dashboard tool was developed by applying the framework to quantify and monitor the change of biophysical condition and forest cover using two common remote-sensing data products that incorporate the Before and After Control Impact (BACI) design (Upton, 2014 and Carpenters, 1989). This approach is used in ecology and environmental studies as an experimental design to evaluate the impact of an environmental change on an ecosystem (Carpenters, 1989). The framework is based on repeated measurements of several vegetation variables at various times over an observational period. Measurements were taken at a location which was known to be unaffected by vegetation change (control location) and at another location which potentially would be affected by this same change (be explicit) (treatment location) for each time step (Upton,2014). The framework is based on the Google Earth Engine cloud computing platform, which is a technology that is able to rapidly deliver information derived from remotely sensed imagery in near-real time.

Recent technological advances have greatly enhanced computational capabilities and facilitated increased access to the public. In this regard, Google Earth Engine (GEE) is an online service that applies state-of-the-art cloud computing and storage frameworks to geospatial datasets.

The framework processes as request data, perform spatial calculations, and serve the information in a browser is shown in Figure 1.

The front-end relies on Google App engine technology. Code developed from either the JavaScript or Python APIs are interpreted by the relevant client library (JavaScript or Python, respectively) and sent to Google as JSON request objects. Results are sent to either the Python command line or the web browser for display and/or further analysis. Spatial information is displayed with the Google Maps API and other information is sent to a console or the Google Visualization API.

Figure 1. The infrastructure for spatial application development provided by Google. The Google Earth Engine consists of a cloud-based data catalogue and computing platform. The Django web server framework is used to host the Earth Engine application.

Methods and data input

Methods

A common method for evaluating the impact of environmental and ecological interventions that applied in this tool development is the BACI (Before-After, Control-Impact) method (Smith, 1993). Figure 2 provides a schematic overview of the BACI framework. For the intervention area, the before and after variables of interest are measured. These are compared with the before and after measures of the same variables at a control site. The differences between the intervention and control sites determine the impact generated by the interventions (Conquest, 2000 and Underwood, 1992).

Figure 2. Schematic overview of the BACI (Before-After, Control-Impact) framework. For both the control and impact sites the before and after situations are evaluated. The difference between the two after situations defines the impact of the measures.

Together with recent advances in cloud-based remote sensing and geo-computational platforms, these technologies have led to greater open scientific data for use by policy makers and practitioners outside of academia. The hybridization and simplification of these technologies also allow scientists to provide policy-makers, international donors, NGOs, and other development partners with tailor-made products to monitor and value their ecosystems in near real-time, and requiring less advanced technical expertise than in the past.

In this tool development, we applied a near real-time method for quantifying vegetation using cloud computing technology and remote-sensing data. We developed a novel custom indicator to monitor vegetation on a planetary scale, and then combined this with remote-sensing data for near real-time customized quantification of vegetation change. We demonstrate how new cloud-based geo-computational technology can be used to temporally, geographically and adaptively filter data collections.

Input data

All data input for the Cambodia Biophysical Monitoring and Evaluation Dashboard are opened and available data on the Google Earth Engine (GEE) that listed below -

Enhanced Vegetation Index

The biophysical monitoring layer uses the Enhanced Vegetation Index (EVI) which is calculated using MODIS multispectral data (MOD13Q1) with a pixel resolution of 250 x 250m. The algorithm chooses the best available pixel value from all the acquisitions from the 16 day period. EVI is a measure of relative biomass and is particularly well-suited for high-biomass areas of the globe.

Tree canopy cover and Tree canopy height

The forest monitoring layer primarily uses annual forest cover time series data. This data is created using Tree Canopy Cover (TCC) and Tree Canopy Height (TCH) layers, which was collated and delivered by the Global Land Analysis & Discovery of University of Maryland (GLAD/UMD) team for the Mekong region (Potapov et al. 2019). The data was taken from time-series analysis of Landsat images characterizing forest extent and change. Using Food and Agriculture Organization (FAO) forest definition with TCC is equal or greater than 10% and TCH is higher or equal 5m, and the minimum mapping unit is 0.5ha.

'Forest loss' is defined as a stand-replacement disturbance, or a change from a forest to non-forest state, from 2000 upto now.

'Forest cover gain' is defined as the inverse of loss, or a non-forest to forest change entirely from 2000 upto now.

Forest alert from the SAR Alert System (SARAS): is a near real time forest alert system using Synthetic Aperture Radar (SAR) the Sentinel-1 satellite image. The system relies on the newest technologies in the field of cloud computing and deep learning classification to identify forest disturbance and deforestation.
Forest fire

The information shows the active fire / hotspot of the Fire Information for Resource Management System (FIRMS). This system was developed to provide near real-time active fire locations to natural resource managers that faced challenges obtaining timely satellite-derived fire information. Data displayed in this tool is the Earth Engine version of the Fire Information for Resource Management System (FIRMS) dataset containing the LANCE fire detection product in rasterized form. The near real-time (NRT) active fire locations are processed by LANCE using the standard MODIS MOD14/MYD14 Fire and Thermal Anomalies product. Pixel displayed as a fire hotspot in this tool achieves 90% confidence of the LANCE fire detection model.

Land cover

The Myanmar National Land Cover time series (2000 - 2023) was developed through a collaborative effort led by SERVIR SEA using the Regional Land Cover Monitoring System approach (Saah et al. 2020). An uncertainty assessment was integrated into the assemblage process. These maps were developed using Landsat annual composites and training data collected by multiple stakeholders.

Components of the LAMP Dashboard

The LAMP Dashboard tools has four components which provides information related to change of biophysical condition and the change of forest cover on the ground, that indicated the area of improved or degraded biophysical condition in the project area; and Annual area of deforestation (ha) and/or reforestation (ha) as the M&E indicator of USAID project requirements. Four components of the tool included: Biophysical monitoring; Forest monitoring, Crop monitoring, and Fire monitoring.

Biophysical monitoring

The Biophysical monitoring component provide information about the change of biophysical productivity of the landscape – scale, uses the Enhanced Vegetation Index (EVI), which was calculated using MODIS multispectral data (MOD13Q1) with a pixel resolution of 250 x 250m. The algorithm chooses the best available pixel value from all the acquisitions from the 16-day period. EVI is a measure of relative biomass and is particularly well-suited for high-biomass areas of the globe.

The map result was obtained by calculating the difference between vegetation growth (positive trend) and vegetation decline (negative trend) over any given area according to the baseline and evaluation period. Negative numbers indicate a net negative trend whereas positive numbers indicate net greening. By tracking changes of the Enhanced Vegetation Index (EVI) user can track the performance of landscape-scale efforts to maintain biological productivity.

A time series of Myanmar national land cover map from 2000 – 2023 is included in this component, that helps users to track the change of vegetation productivity with the land cover change during the evaluation period quantitatively.

Forest monitoring

The forest monitoring component provides information about the annual forest cover and forest cover change during the evaluation period. The map of annual forest cover and map of forest gain and loss during the evaluation period was created using the annual layers of Tree Canopy Cover (TCC) and Tree Canopy Height (TCH) from GLAD lab of University of Maryland (UMD) for Mekong region, using Landsat data with the forest definition of FAO (10% of TCC and TCH above 5m). In this component, users can access a time series data set of annual forest cover and forest cover change, with total area of forest gain and loss comparison between baseline and evaluation period. These data can help users to track the result of the landscape enhancement activities, and provide quantitative information about the area of forest change.

Crop Monitoring

The crop monitoring component provides information about map of seasonal rice areas for two periods 2017-2020 and 2021-2023. The maps from 2017-2020 are derived from radar imagery captured by Sentinel-1 (10m), while the maps from 2021-2023 incorporate additional optical imagery from Sentinel-2 (10m) and Planet-NICFI (5m). These maps are the output of a random forest model, run in Google Earth Engine, which estimates the probability of each pixel being rice or non-rice. From Monsoon 2021 onwards, rigorous validation following Olofsson et al. (2014) is carried out in the Collect Earth Online (CEO) platform. Area estimates are either ‘pixel-based’ as outputted by the model or ‘sample-based’, representing unbiased area estimates at a state/region level.

Forest alert
The forest alert component provides information about map of forest disturbance alerts with both SAR alert and GLAD alert data by year. SAR Alert system is developed using Sentinel-1 data applied deep learning method in TensorFlow. The system is operated as part of the Cambodia Protected Area Monitoring Platform (PAMP) which can monitor near real-time forest changes and external threats including deforestation, forest fires and floods within the extended Prey Lang Wildlife Sanctuary. Which integrates ground-based and remotely sensed data to improve the efficiency and effectiveness of law enforcement and conservation activities.
Fire hotspot

The Fire hotspot component provides information and a map of fire hotspots of the defined area or landscape by year. The information shows the active fire / hotspot of the Fire Information for Resource Management System (FIRMS) data. This system was developed to provide near real-time active fire locations to natural resource managers that faced challenges obtaining timely satellite-derived fire information. Data displayed in this tool is the Earth Engine version of the Fire Information for Resource Management System (FIRMS) dataset containing the LANCE fire detection product in rasterized form.

SERVIR South East Asia | ADPC