Water Monitoring of the Murray-Darling Basin Using Time Series Data
Exploratory analysis of historical data collected along the Murray-Darling basin.
One of the most important agricultural regions in Australia is irrigated by a large system of rivers, known as the Murray-Darling basin. The system takes it’s name from the two main rivers, the Murray and the Darling, whose combined length snakes through more than 3000km of land and five Australian states. Monitoring the water flow, as well as other water quality measures, will support improved agricultural activities within the basin.
Water management is of increasing importance for the river system due to climate change, increasing water usage, ecological damage, and political uncertainty around the Murray-Darling Basin Plan. Better understanding and prediction of the river will enable us to better manage the water for those reliant on the basin. For example, if we can better predict and share water flow and salinity 2–4 weeks in advance, we can change irrigation and usage behaviour to benefit both agriculture and other users of the river. With this in mind, I have conducted some exploratory data analysis of Murray-Darling data sets.
Amphora Data from the Murray river at Albury
I explored historical data on Amphora Data, a platform to discover and share data. They have free historical and real-time data across the Murray-Darling basin in machine readable format. I have started this data exploration with data collected from the Murray at Albury, on the Victoria-New South Wales border. In the future we will look into extending the research by combining and comparing the analyses from several locations along the basin.
The dataset contains the following water measurements:
- level (in meters, 1885 to 2019);
- temperature (in degrees Celsius, 2001 to 2019);
- salinity (in microsiemens per centimetre, 2010 to 2019).
Analysis setup with RStudio and Kozai
The exploratory data analysis was performed using RStudio via the Kozai platform. In particular, I used the R package feasts which contains a suite of extremely powerful tools to visualise time series data organised in the tsibble data structure. This was complimented with tidyverse packages such as ggplot2 and dplyr. The code used to produce the visualisations in this post is available on the Eliiza GitHub account.
Water level at Albury over the last century
The visualisation of the time series below shows the absence of a clear trend over the period. A striking behaviour at this resolution is a sharp drop in the water level slightly prior to 1900. We are currently uncertain as to the causes, however it could be attributable to the construction of a dam somewhere upstream along the river. It is also possible to observe unusually low levels between 2001–2007 (highlighted in red) which corresponds to a long period of drought for South-East Australia.
Seasonality of water level and temperature
A closer look at the characteristics of the river after year 2000, shows obvious seasonality in water level and temperature. While extremities were identified in the salinity range in 2015 (highlighted in red), this was not reflected in the other two measures.
The below interval breakdown of the water levels highlights an overall consistency in the seasonal patterns over the century. Each line in the plots represents the time series for one year. The highest levels are observed roughly between July and December, even though this period has been shortening over the century. Similarly, the period with the lowest levels has been expanding and showing less deviation from the average. This pattern could be attributed both to environmental conditions (rainfall, temperature etc..) and to the amount of water used for irrigation, which has been increasing over the century.
The yearly breakdown of water temperature has a strong seasonal pattern, reaching its highest levels during summer (January to March). While the regular seasonal pattern makes it a predictable measure, the warmest months have higher variability, showing larger deviations from the average trend. Interestingly, years 2008/2010 and 2017/2018 show temperatures significantly higher and lower than the average respectively.
Extreme water salinity or technical problem?
Unlike temperature and level, water salinity lacks a remarkable seasonal pattern. The coldest months (June-August) generally show higher variability which correspond with lowest water levels. Surprisingly, salinity in 2015 registered unusual levels, with a steep increase in the month of January. The sharp unusual change over several weeks introduces the possibility of technical problem in data collection. It is important to clarify whether this was due to a technical fault, as it might impact the reliability of future predictions.
Correlation of water level with salinity and temperature
Finally, I explored the relationship that water level has with the other two available measures. In the plot below, the measurements of the water levels are plotted against the corresponding water salinity. A smoothed line has been fitted to the data and demonstrates the absence of a correlation. The only exception is 2015 (in red), showing a positive association between the measurements, i.e. higher salinity at higher water levels. However, caution must be taken in drawing any conclusions on this year as the salinity analysis highlighted this year as a potential anomaly. In addition, the scatterplot shows that the variability of the salinity decreases as the water level increases.
To explore the association between water temperature and level, I produced a heatmap of hexagonal bins. The R package hexbin was used for this and lighter colours indicate high density. This visualisation shows an initial positive association between the measurements, with lower temperatures observed at shallower water levels (< 1.5m) and higher temperature at higher levels. However, the temperature decreases again after approximately 2m. Indeed, the highest water levels are observed from the middle of winter until the end of spring which are usually characterised by low temperatures.
Final remarks and future directions
This exploratory data analysis shows the seasonality observed in monitoring water information collected at Albury, along the Murray river. The analysis discloses the potential application to farming and land management in predicting water level and temperature as well as monitoring extreme levels of salinity. In particular, the salinity of the water is an important parameter to be kept within appropriate thresholds to prevent damage to the crop and land, as high levels of salt in the soil tend to reduce growth.
An interesting extension to this analysis could be to combine and compare other Amphora land and weather data collected at different locations within the Murray-Darling system. This could show whether similar patterns are observed and if there are differences between the upstream and downstream characteristics of water usage. Proper modelling of the time series data would provide real-time predictions for water characteristics in the future. Overall, this has the potential to facilitate a deeper understanding of water usage and sustainability of the largest resource for irrigation in Australia.
