Geoimage uses DigitalGlobe technology to develop a machine learning solution addressing one of Australia’s most significant environmental challenges – the Murray-Darling Basin

Editor’s Note: The DigitalGlobe 2018 Australia Sustainability Hackathon aimed to address Australia’s most conflicting issues surrounding mining, agriculture and environmental sustainability using machine learning and satellite imagery. This blog post is written by the winning team from the environmental sustainability category.

The Vast and Complex Murray Darling Basin

The Murray-Darling Basin (MDB) in Australia is a vast catchment covering more than 1 million kmand with a primary water course exceeding 3000km in length. It is the life blood of many rural and regional communities and supports the food bowl of the nation. To put the scale and complexity of the MDB into perspective, the basin area is equivalent to the 30th largest country on the planet, it traverses 5 states and it drains about 15% of the Australian land mass.

In 2014, concern about the deteriorating condition of the basin’s ecosystem and subsequent impact on flora, fauna, agriculture, industry and communities led to the formation of the Murray-Darling Basin Authority to oversee this valuable resource. The economic, environmental and social consequences cannot be overstated.

Geoimage Team Responds to the Challenge

The DigitalGlobe 2018 Australia Sustainability Hackathon proved to be the perfect motivation to inspire the Geoimage Team to imagine how the power of DigitalGlobe’s Geospatial Big Data platform (GBDX) could be used to solve a complex problem covering vast areas in an efficient and reliable manner.  The Team challenged itself to develop a repeatable, broad-scale measurement of the MDB’s health to enable stakeholders to make more informed decisions. For the purposes of the Hackathon, the Geoimage Team focused on 2 critical health metrics – water quality (turbidity) and changes in water volumes.

 Satellite Imagery and GBDX Provide the Smart Data and Analytical Platform

Comprehensive, repeatable data capture over large areas at varying points in time is a major benefit of using satellite sensors for this type of research. With high-resolution imagery and the recent emergence of big data analytics on platforms such as GBDX, the power to perform studies on a massive scale at repeatable intervals is now possible.

 Lake Albert (South Australia) – Proof of Concept Study Area

As proof of concept, Geoimage used GBDX to focus on a small study area within the MDB with the theory being that, if successful, this process could then be applied across larger areas of the MDB.

Our study focused on Lake Albert, a 170km2 water body located 100km southeast of Adelaide near the mouth of the MDB river system.

MDB Catchment Area >1Mil km2 SE Australia

Lake Albert Study Area Methodology – Changes in Water Volumes

To begin our study, the team located data from 2008 – the point in time when the drought was at its worst and the lake volume was at its lowest. We found Digital Elevation Model (DEM) data from 2008 and then retrieved imagery from 2008, 2014 and 2018 to derive water masks using the Normalised Difference Vegetation Index (NDVI). The intersection of the DEM and each water mask provided a way to calculate the change in water volume on a pixel by pixel basis, relative to 2008.

The figures below show the 2008, 2014 and 2018 water masks that delineate Lake Albert.

The volume change results are shown in the graph below, reflecting a significant 15 gigalitres (GL) increase in Lake Albert’s water volume between 2008 and 2014 and indicating significant recovery following the end of the drought in 2011. Between 2014 and 2018, there has been only a minor 0.7GL increase in volume, reflecting a more stable period of in and outflow.

Lake Albert Study Area – Change in Water Volumes 2008 – 2018

Lake Albert Study Area Methodology – Water Quality (Turbidity)

Suspended material within bodies of water can have a detrimental impact on aquatic life. Measuring levels of this material, in addition to water volume, provides a more robust indication of the health of the basin.

In this study, we derived several different indices over Lake Albert using imagery captured between October 2017 and February 2018 (a period spanning the ‘wet season’ in central and northern Australia when heavy rains flow from the rivers southward). The figures below illustrate the amount of suspended material, demarked by a more prominent yellowish discolouration in the channel leading from the larger Lake Alexandrina (top left corner of the image) into Lake Albert. The extent of the plume extends further into the body of the lake as the wet season progresses.

Images above: Amount of suspended material in Lake Albert between Dec. 2017 – Feb. 2018, with February reflecting the heaviest presence.

Successful Proof of Concept Warrants Further Analysis of the MDB

While our study focused on a relatively small area of the MDB, the results show potential for informing authorities of the health of the entire basin. Supplemented with other datasets such as rainfall, climate, groundwater, land use and water allocation, GBDX and high-resolution imagery can be used to build a model that provides real insight for scientists, environmental managers, farmers and policy makers alike.

A Winning Team Effort – Geoimage and DigitalGlobe’s GBDX

The study proved that the winning combination of Geoimage’s geospatial analytical skills and DigitalGlobe’s satellite Imagery and GBDX can deliver an effective, high-value solution to one of Australia’s most significant environmental, economical and social challenges.