1. Introduction
Although common carp (
Cyprinus carpio, hereafter referred to as carp) was originally introduced in southeastern Australia in the nineteenth century, it was not until the 1970s that it became widespread and by the 1980s it was accepted as a serious invasive species [
1]. This recognition followed surveys which showed that, in some parts of the Murray–Darling Basin (MDB), common carp comprised over 90% of the fish biomass with an associated loss of native fish species populations [
2]. Nevertheless, carp do not dominate in all hydrological ecosystems in southeastern Australia, and the relative importance of river regulation versus the capacity of carp to be ecosystem engineers, as well as the processes by which they affect water quality are topics of ongoing debate [
3,
4,
5].
Whilst the role of carp as a cause or a consequence of ecosystem decline in the MDB is contentious, less so is the need to suppress the population of carp to achieve native fish recovery [
6]. Accordingly, a number of options have been explored over the past 30 years [
1], and a National Management Plan was adopted in 2000 [
7]. Despite these, no wide-area control has been achieved, with commercial harvest, selective poisoning, genetic control, and physical separation proving either to be ineffective or cost-prohibitive [
8].
On the basis of Australia’s success using the Myxoma virus to reduce an invasive rabbit population, spring viremia of carp virus (SVCV) was proposed as a potential biocontrol agent for carp [
9]. However, subsequent research has found that the virus was not specific to carp (Fam.
Cyprinidae) and even affected non-cyprinid fish such as sheatfish (
Siluridae), guppy (
Poeciliidae), and northern pike (
Esocidae) [
10]. Subsequently, viral biocontrol was rejected as an option, and the focus of research became the potential use of sex-biasing genetic modified carp (“daughterless carp”) [
11]. Unfortunately, this technology did not fulfil its early promise, at which time viral biocontrol became a more feasible option, due to the detection of cyprinid herpesvirus 3 (CyHV-3) as a cause of mass mortality events in carp in the late 1990s [
12]. Subsequently, the virus spread to numerous carp rearing countries, including Indonesia [
13], from which an isolate was transferred to Australia’s high quarantine animal health facility, the Australian Animal Health Laboratory (now the “Australian Centre for Disease Preparedness”). Subsequent research confirmed that, unlike SVCV, CyHV-3 was very specific to carp, and did not infect native Australian fish [
14].
A feature of many viral infections in fish, including SVCV and CyHV-3, is that infection and disease only occur naturally in a defined water temperature range [
15]. For CyHV-3 infection in carp, this range has been established by infection trials to be between 16 and 28 °C [
16,
17,
18]. The existence of a “permissive range” of temperature potentially limits CyHV-3 as a biocontrol agent, particularly if rivers and waterways are only within this range for a short period, or else the water temperature oscillates near the upper or lower threshold, such that infection might not progress to disease. Indeed, a method proposed in aquaculture to immunise carp against CyHV-3 involves infection within the permissive range, and then raising it above this to prevent the appearance of disease [
19].
While many factors can influence the complex interaction between the virus and the fish to produce disease, such as CyHV-3 persistence in the natural environment and its capacity for latency [
20], water temperature is seen as the most critical driver for a successful virus release strategy. A simple assessment of the potential of CyHV-3 with respect to water temperature has been undertaken using a dataset from a single point beneath a weir in New South Wales, and it has confirmed that the water temperature was within the permissive range for most of the spring, summer, and autumn [
21]. However, the generality of this conclusion across the rivers of southeastern Australia is yet to be substantiated, and in particular, whether it applies to intermittently flooded wetlands, which are important for carp’s invasiveness on account of enabling massive recruitment events [
22]. Unlike the main river channels, very little systematic water temperature data have been collected, because of the often transient nature of water temperature [
23]. Furthermore, because of the complex hydrology of the MDB, it is not possible to simply use air temperature to estimate water temperature, as has been applied in countries with more stable hydrology [
24]. Thus, there is a need to integrate air temperature measurements with flow estimates, while also considering the nature of the waterbody [
25].
Therefore, the objective of this study was to model water temperature in both rivers and waterways, to a high degree of precision over an extended period. Due to the size of the MDB, we restricted the study to five catchments with a total of 132,129 km2 of drainage area which was representative of the diversity of freshwater environments and the majority of carp habitat found in the basin. A second objective was to assess the extent to which water temperature modelling could be used as the basis for operational planning of the timing of release of the virus, and therefore maximize its activity. For this planning, we reasoned that the ideal would be if the waterbodies entered the permissive range reliably each year at a certain week during the spring warming and remained within it for a period sufficient to readily transmit the virus among the fish. To assess the interannual variability, therefore, we needed to estimate the water temperature over an extended period of time. For further studies, by integrating hydrological and temperature controls into detailed demographic and epidemiological models of carp populations and virus spread, we provided reconstructed hydrological landscapes, incorporating river flow, inundation, and hydrological connectivity together with water temperature. Finally, the results should support the development of a release strategy for CyHV-3 for common carp control in southeastern Australia
4. Discussion
Defining constraints for carp herpesvirus release in a large connected river basin needs the interplay of various modelling and computational techniques to be successful. Large scale hydrological models are available for parts of the Australian system, or worldwide. The status of connectivity between water bodies over large regions can only be achieved by using remote sensing technology. Connectivity might be a lesser issue in regions with regular flow characteristics but is an essential part of flood inundation and disease spread modelling in slow-flowing, lowland river systems. The high variability of climate conditions in Australia, which drive large variability in flow over seasons and interannually in Australian river systems, made it necessary to include a dedicated component of flow and connectivity modelling for our task. Going a step further, we included a region-wide model for water temperature simulation based on basic meteorological data readily available for the whole southeast Australian region, the size of France and Germany together, or one-third of the entire Mississippi basin. This unique combination of large-scale hydrological reconstruction should enable essential input to facilitate further modelling of carp habitats, population dynamics, and epidemiological modelling to predict how CyHV-3 might spread across the hydrological landscape and result in carp population suppression.
Our reconstruction of the hydrological environment over such an extensive area and period at a fine spatiotemporal scale and its use in the subsequent integration with carp habitat and virus epidemiological models was only made possible by explicitly adopting a big data approach, i.e., rapidly processing and integrating large volumes of heterogeneous data for decision making. The determining criteria for big data are volume (amount of data), velocity (speed of data processing), and variety (different data types) [
71]. Other “v” terms are veracity (removal of erroneous data by error trapping processes), validity (replicability and quality assurance of data management and processing), and value (a big data system needs to be useful). Applying these concepts to our reconstruction of the river and waterway environments, it was necessary to handle a moderately large volume (at approximately 1.8 TB) of very diverse data types and formats, the latter ranging from raster satellite imagery to stream segments to time series of flows. Due to the anticipated importance of the results from the modelling (i.e., the value), much effort was needed to make the science transparent and replicable (i.e., valid and veracious). To achieve this, processing was coded and input/output for all steps stored within a scalable PostgreSQL database with regular backup and retrieval systems. The only big data requirement not particularly important was processing speed, although in practice a cloud computing infrastructure was used for all runs.
In practice, the greatest challenge faced for the river and waterway environment reconstruction was the availability and quality of the input data. In particular, this applied with the hydrology for the rivers and streams away from the main channels (for which there is in general little flow gauge data) and especially for the non-Murray River catchments. Thus, for example, whilst it was possible to obtain quality flow data for the main channel of the Lachlan River catchments for the entire study period, for the tributary rivers and streams, it required rainfall-runoff modelling be undertaken, and quality data arising from this was only available from 2000 onward.
Similarly, to estimate the timing and area of inundation of wetlands and floodplains, whilst, for the Murray River, we could use the output from the existing RIM-FIM inundation model, for the other catchments, for which this modelling has not been applied, we needed to rely on satellite imagery. Using this, we found a number of inconsistencies when the imagery from Landsat TM and MODIS were compared, due in part to their different spatio-temporal resolution, i.e., Landsat’s 16-day frequency could not pick up highly ephemeral waterbodies, whilst MODIS, with a spatial resolution of 500 m resolution, could not detect small permanent ones [
53]. An additional problem of using satellite imagery for estimating inundation is in detecting water presence in highly vegetated areas such as the Great Cumbung Swamp in the Lachlan River, as the overlapping of vegetation and water within a pixel misclassifies the pixel to vegetation and not swamp or highly vegetated wetlands [
40]. Therefore, it is probable that we underestimated the extent of inundation in this area, as compared with the Barmah-Millewa Forest in the mid-Murray, where the inundated areas were estimated by the more precise RiM-FIM modelling. Even then, the RiM-FIM model can only estimate the area based on the flows of the associated river gauge, and in some areas, such as Lake Victoria on the lower Murray River or Lake Moira within the Barmah Forest, when flows are below the commencement-to-fill value and yet there is standing water, the predicted water area may not be accurate.
Water temperature is an essential parameter in developing models of the potential behaviour of CyHV-3 in natural populations of common carp. Water temperature can it have a direct effect on virus replication within infected fish, with the permissive range generally considered to be between 16 and 28 °C, and it also has strong effects on seasonal reproductive spawning aggregations, and thereby recruitment of susceptible juveniles into the population. Furthermore, water temperature is an important factor in the habitat suitability of rivers and wetlands for carp, and thus their population density. Here, we established a general method for a landscape-level reconstruction of the hydrological environment (1990–2017) across five catchments in southeast Australia to define water temperature constraints for the release of the biocontrol agent, cyprinid herpesvirus 3 (CyHV-3) to control common carp (
Cyprinus carpio). The water temperature simulated here for the five catchments can be used to set up a physical-based release strategy considering the diversity in water temperature on a north-south gradient and within the catchments over a seasonal cycle, as well as flow and connectivity between water bodies. As water temperature readings from gauges in this large system were sparse and hardly available for periods before the year 2000, we used the available temperature data to set up water temperature models driven by air temperature and flow [
66,
67] for specific locations and generalising them for the entire catchments and for the whole time period. This approach is easily scalable to the entire southeast Australian region or even continent wide.
In deep lakes, water temperature does not vary as much laterally as with depth, leading to a different type of habitat separation. Stratification is persistently present in deep lakes, and shallow lakes or river reaches might show non-persistent stratification during warm spells. This behaviour was simulated using a hydrodynamic model accounting for the full heat balance [
62,
63], requiring additional computational resources and a more detailed database on local meteorological data, as well as continuous water temperature recordings at multiple depths, which, in general, were not available for most deep lakes or reservoirs in Australia. Furthermore, the release of cold bottom water from reservoirs can lead to downstream cold water pollution [
69,
70], which must be considered when developing a virus release strategy.
In general, water temperatures of Australian rivers and waterways are within the permissive range for CyHV-3 activity for periods in spring, summer, and autumn, which is in agreement with postulates from [
21]. However, these periods vary in extent and occasion, depending on their geoclimatic position (north-south, altitude). The time of year when this starts and ends is highly variable between and within catchments, with strong latitudinal and altitudinal gradients being evident. The interannual variability can be large, and any release strategy must determine the right timing between different regions to be most effective. Becker et al. [
21] only examined four upstream river reaches in a very confined region of the basin, each of which was downstream of a weir or dams, and thus possibly affected by cold water pollution which, to some extent, is common, but not a general feature for the waterways in southeast Australia. They concluded that, for those very limited examples, the permissible range of virus activity was met for large periods of the year. Here, we show that the picture is much more complex and must differentiate between catchments in different climate zones, allow for local effects such as being downstream of dams, as well as seasonal and interannual variability in connectivity between water bodies. A simple look at water temperature in a single region would bias the conclusions of viability and effect of a virus release. By contrast, through examining water temperatures across catchments in different climatic regions in southeast Australia, we provide an overview of possible periods of opportunity for virus release depending on water temperature, as well as flow and connectivity. This big data approach yields a much more varied picture. It shows that a virus release strategy must be accompanied with detailed hydrologic and climate studies in the catchments to cope with the large variability of climate, water temperature, and flow characteristics throughout southeastern Australia.
Whilst we show that the water temperature is generally permissive for virus activity during extended periods in the spring, summer, and autumn in the rivers and waterways of southeastern Australia, nevertheless, it is premature to conclude that the virus would actually result in carp mortalities. Thresher et al. [
72] collected data on fish kills from North America associated with CyHV-3 and showed their occurrence in natural populations only during spring. The same was evident in Japanese records of CyHV-3 outbreaks, where although autumn outbreaks occurred, these were mainly in aquaculture farming [
73]. The predominance of outbreaks in spring in wild carp populations in Japan has been hypothesised to result from the direct contact which occurs during the spring spawning period [
74]. This suggests that in order to predict the impact of CyHV-3 on carp populations in southeastern Australia, there is a need for a fuller understanding of the demographic structure of carp over seasons and the clarification of habitat structure in the basin to determine hotspots of fish aggregation during spawning.
5. Conclusions
Using different model tools for streams and lakes, we were able to set up a unique, first of its type, model to describe waterbody connectivity, flow, and water temperatures across five catchments over an extended area of southeastern Australia (>130,000 km2). The choice of models was driven by integrating available gridded meteorological data, gauged and modelled flow data, and remote sensing imagery. We did not aim to model individual river reaches, wetlands, or lakes including all local characteristics, for example, along a shaded reach, or cold water pollution downstream from a large dam. The model system was integrated into a database system which embedded a big data approach capable for generalising it across the entire southeast Australian region.
The results of the hydrological reconstruction across five distinct regions in southeast Australia highlight the large variability in connectivity, flow, and water temperature in both space and time. This variability leads us to conclude that a small-scale approach in terms of spatial and temporal coverage cannot be used to give a general answer to timing, location, and staging of a CyHV-3 release across the wider region or Australia. Furthermore, the Northern Hemisphere experience of outbreaks occurring in wild populations, predominantly in the spring, suggests that water temperature modelling alone, cannot be used as a basis for developing a strategy for the optimum release of the virus. Thus, to achieve this goal, there is a need for integrated modelling of the biotic factors affecting carp populations, such as movement, reproduction and recruitment, and how these might interact with the epidemiology of the disease induced by the virus. We conclude that, whilst temperature modelling is certainly essential for developing a release strategy for CyHV-3, it is not by itself sufficient, and further integrated modelling is required.