Cultivating Success: Cambodia’s Crop Mapping

Powered By Tensorflow And Google Earth Engine

August 22, 2024
7:06 pm

share this post:

Following the creation of the Paris Agreement, the Food and Agriculture Organization of the United Nations (FAO) created the Capacity Building Initiative for Transparency (CBIT Project), intending to support developing countries in meeting transparency requirements while accurately monitoring progress toward their Nationally Determined Contributions. As an element of the CBIT Project, the FAO enacted measures to assist the government of Cambodia with reporting progress to the United Nations Framework Convention on Climate Change, a task made difficult by the country’s lack of efficient agriculture and land-use data collection systems. To help in this endeavor, Spatial Informatics Group (SIG), in collaboration with SilvaCarbon and the FAO, started a project on developing a crop mapping programme in Cambodia, conducted by a team of the General Directorate of Agriculture (GDA). SilvaCarbon is an interagency technical cooperation program of the US Government to enhance the capacity of selected tropical countries to measure, monitor, and report on carbon in their forests and other lands. US Forest Service and US Geological Survey are leading the implementation of SilvaCarbon. SilvaCarbon has been supporting the Cambodia crop mapping initiative since 2021. At the core of this project is compiling a collection of crop mapping tools and providing training on how to apply those tools to expand available land-use data and to develop and validate unbiased area estimates of crop maps.

To ensure that the initiatives developed adequately addressed Cambodia’s needs, SIG teams worked closely with the Cambodian Department of Agriculture and Land Resource Management (DALRM), identifying gaps in the existing data and developing a remote sensing-based approach to improve existing methods of crop mapping. The initial consultation between SIG and the DALRM revealed the need to emphasize the mapping of economically valuable crops, such as cashews and mangos, in order to increase the productivity and profitability of Cambodia’s agricultural sector, as well as to more accurately report agricultural greenhouse gas emissions. In response to user need, a team from Spatial Informatics Group (SIG) developed a machine-learning workflow based on Google’s software library TensorFlow that is capable of analyzing satellite imagery to create detailed crop maps at a fast pace and low cost. Turning to the Google Cloud Platform, SIG team members trained Cambodian officials in effectively using Google Earth Engine tools to conduct geospatial data analyses and access libraries of helpful satellite images and data. After establishing a more efficient crop mapping system catered to the Cambodian government’s needs and implementing the use of Google Cloud storage for finished mapping models, SIG’s team members provided consistent support in the development of new crop mapping models using Keras, a TensorFlow-based library. Online training, in-person meetings, and weekly consultations with members of the DALRM were coordinated. SIG also provided capacity building training intended to help users become familiar and confident with machine learning, specifically with Neural Network Models on Keras, and created a training website of detailed instructions for various crop mapping tools. Said Chaya Veasna, a member of the DALRM, “Google Earth Engine has been very valuable for our office at the Cambodian Department of Agricultural Land Resources Management. Its powerful features enable us to monitor forests, analyze land use, and generate agricultural maps, including the recent achievement of mapping our national cashew cover.

Before the introduction of this cloud computing workflow in GEE, it took our team a year to update the land cover map for just one province. After our team of four were trained on this workflow, we were able to map cashew plantations across 25 provinces, approximately 400,000 hectares of this important crop.”

SIG trainers, Andréa Puzzi Nicolau and Ate Poortinga, during the Machine Learning/Google Earth Engine training workshop in Kampong Thom, June 2022.

In the two years since SIG, SilvaCarbon, and FAO began working with Cambodia’s DALRM, the remote-sensing based crop mapping models developed have been successful in improving the Cambodian government’s ability to monitor and report progress against its Nationally Determined Contributions. Additionally, new data obtained on agricultural land-use will bolster Cambodia’s ability to adapt to climate change and confront agricultural greenhouse gas emissions. The partnership between SIG and the Cambodian DALRM is set to continue with increased emphasis on assisting DALRM members in utilizing Teras Neural Network models to create area and uncertainty estimates of crop maps, ultimately cultivating Cambodia’s overall sustainable development.

Cashew (left) and land cover (right) in Kampong Thom Province, Cambodia.

Cultivating Success: Cambodia’s Crop Mapping

Powered By Tensorflow And Google Earth Engine

Info

Science Groups

Law