Waterbody scale assessment using spatial models to identify suitable locations for cage aquaculture in large lake systems: A case study in Volta Lake, Ghana

Large lakes throughout the world offer development opportunities for cage aquaculture. However, due to their vast area, identifying the most suitable locations can be a challenge. This is also complicated as lake systems are often multi- use environments, so a strategic approach to managing the resource is required. This study uses geographic information systems (GIS) to develop a broad- scale approach that identifies potential areas that may be suitable for cage aquaculture development. Volta Lake, one the world's largest man- made lakes, is used as a case study. The overall GIS model combines four sub- models, bathymetry, hydrography, water quality and access, and a constraints layer, to identify the most suitable locations for tilapia production. Three different cage sizes are modelled: small, medium and large. The model outputs suggest that approximately 102 km 2 (1.7%), 406 km 2 (6.9%) and 407 km 2 (6.9%) of Volta Lake can be categorized as highly suitable for development of small, medium and large cages respectively. A further 634 km 2 (10.8%), 1264 km 2 (21.4%) and 1055 km 2 (17.9%) can be categorized as suitable for the same. The results can be used by stakeholders and decision makers to identify specific areas where aquaculture development for cage farming of tilapia could be prioritized.

SDGs, there is a need to identify opportunities to increase food production sustainably and develop strategic plans to maximize benefits and minimize negative impacts.
Aquaculture is an important food production sector, and if planned and managed appropriately, it can provide safe and nutritious food as well as socio-economic benefits for local and national communities (Soto et al., 2008). Freshwater cage aquaculture in large lake and reservoir systems is recognized as a way of increasing fish production that would contribute to food security and help achieve the SDGs if carefully planned and managed (Cowx & Ogutu-Owhayo, 2019). A key part of this is identifying the most suitable locations for development of aquaculture. Lakes are used for many different purposes (Reynaud & Lanzanova, 2017), and there are often areas that will not be available for development. Constraints, including physical barriers, legislative restrictions or other resource users, will limit opportunities to establish cage aquaculture sites.
Furthermore, even if areas are available for development, they may not be suitable. An aquaculture site must fulfil certain biological, environmental and socio-economic criteria if a farm is to be established . Identification of suitable areas for production is a prerequisite of sustainable development.
For large lakes, the size of the waterbody, which can extend across hundreds or thousands of km 2 , presents a challenge for identifying suitable areas for aquaculture development. Conditions can vary considerably throughout a lake system, but it is costly and time-consuming to perform detailed fieldwork and environmental sampling. Planners and aquaculture producers require more costeffective solutions to identify priority areas where they should target resources and focus their attention. Geographic information systems (GIS) can be used to combine multifactorial data to produce an output that helps decision makers identify potential locations that are likely to be suitable for aquaculture based on defined criteria (Falconer et al., 2020). Stakeholders can then focus on these areas for more detailed analysis through data collection and modelling to establish whether these locations are suitable for aquaculture or not. The use of GIS and spatial modelling for aquaculture planning and management has been used in many studies (Aguilar-Manjarrez & Ross, 1995;Falconer et al., 2013;Longdill et al., 2008;Pérez et al., 2005;Ross et al., 2011;Salam et al., 2003). However, to date, there have been far fewer applications of spatial modelling for freshwater cages compared with other aquaculture production systems (Falconer et al., 2020).
Large lakes throughout the world would benefit from more strategic planning of freshwater cage culture and a GIS-based approach to help identify suitable areas for production can facilitate this. In Africa, cage aquaculture is growing rapidly in natural and man-made lakes (Musinguzi et al., 2019), in many cases prompted by a decline in capture fisheries (Cowx & Ogutu-Owhayo, 2019;Temesgen et al., 2019). Volta Lake in Ghana, West Africa, is one of the world's largest man-made lakes, and although primarily created for hydroelectric power generation, aquaculture has become an important economic activity (Amenyogbe et al., 2018). A key objective of the Ghana National Aquaculture Development Plan is to increase output from aquaculture to make up for the short fall in domestic supply from capture fisheries (Fisheries Commission, 2012). Concerns have been raised about the potential impacts of increased cage tilapia production on the ecology of the lake and its other uses (Ameworwor et al., 2019;Asmah et al., 2014). Furthermore, Rurangwa et al., (2015) reported 1200 cages were recently abandoned by producers. There are a number of factors behind cage abandonment, but the main factor was found to be fish mortality, thought to be a result of low oxygen levels, poor water quality and disease outbreaks (Mantey et al., 2020). Inadequate site selection considerations may have contributed to these problems. The process of siting cages in Volta Lake lacks a formal structure and needs strategic guidance to maximize potential sustainable production. If cage aquaculture is planned appropriately, Volta Lake, and other large lakes worldwide, can play an important role in increasing fish production and contribute to achieving the SDGs.
The aim of this study was to develop a GIS-based approach to identify suitable areas for tilapia cages in Volta Lake that can be prioritized for more detailed assessments to support site selection and sustainable development of aquaculture. Though Volta Lake is the focus of the study, the approach and findings are relevant to other large lake systems in Africa and globally. The work demonstrates the usefulness of broad-scale GIS assessment as a cost-efficient planning tool for aquaculture in great lakes.

| Study area
Volta Lake lies between longitudes 1°30′W and 0°20′E and latitudes 6°15′N and 9°10′N (Figure 1). It is one of the largest man-made lakes in the world and at the maximum level has a volume of 149 km 3 , a surface area of about 8500 km 2 , mean depth of 19 m and a length of 500 km. It is dendritic in shape and forms part of the Volta Basin which has a catchment area of approximately 394,000 km 2 shared by six countries namely Mali, Benin, Togo, Burkina Faso, Ivory Coast and Ghana. The lake has been divided into eight strata to facilitate fisheries stock assessment and management (Béné, 2007;Vanderpuye, 1982) (Figure 1). These strata have been used extensively for biological, limnological and economic studies of the lake.
Here, they are used as a basis for hydrographic and water quality monitoring.
The climate type of the basin is tropical continental or savannah, with a single rainy season in the northern section which spans July to September, and two rainy seasons in the southern part of the lake with peaks in June/July and September/October. These are followed by a prolonged dry season. The annual rainfall ranges between 1000 and 1150 mm. Inflows to the lake are largely influenced by discharges from the major tributaries which are all influenced by climatic conditions in northern Ghana.
Besides fishing by the riparian communities, agriculture is the major land use activity in the basin, with a large percentage of the inhabitants being engaged in extensive cultivation of crops and livestock rearing. Cage fish farming sites are presently concentrated in stratum II.

| Model development
The spatial model was developed using the GIS software terrset The classification of suitable areas for cage aquaculture zones on Volta Lake was based on a constraints layer (areas that are unavailable for aquaculture) and four sub-models: bathymetry, hydrography, water quality and access. In each sub-model, the data layers were reclassified to the same scoring system where 1 is considered highly unsuitable and 5 is highly suitable. Similar studies have used other scoring systems, such as a 1-4 scoring system developed for southwestern Bangladesh (Salam et al., 2003); 1-8 scoring system used in Tenerife, Canary Islands (Pérez et al., 2005) and a 1 to 15 scoring system used in the State of Sinaloa, Mexico (Aguilar-Manjarrez & Ross, 1995). For the present study, a 1-5 scoring system, as used by Ross et al., (2011) for a similar site selection study in Mexico, was selected as it allowed sensitivity but was not too complex to implement (Table 1). Scores were assigned to data in the layers based on values from the literature and/or expert opinion.
Producers operate a range of cages of different sizes and net depths in differing numbers and configurations. Site suitability is based on the cage sizes and configuration rather than on size of the operation. Consequently, three cage sizes representative of systems which are, and could potentially be, used in Volta Lake, were selected for the model: small cages (5 × 5 m, net depth of 4 m plus 2 m for net hang); medium cages (15 × 15 m, depth of 5 m plus 2.5 m for net hang) and large cages (circular cages of F I G U R E 1 Volta Lake, Ghana, showing the water quality (red) and current (blue) sampling locations. The lake strata are shown (green). Stratum I = Afram arm; II = lower main body of the Lake; III = middle main body; IV = upper main body; V = Oti river arm; VI = lower Volta riverine body; VII = middle Volta riverine body; and VIII = Upper Volta riverine body (Vanderpuye, 1982) (Karikari, 2016) and allows decision makers to assess the suitability of the lake for three different scales of aquaculture. The small cages were considered suitable for subsistence or local market farming, while the medium and large cages are more suitable for commercial operations seeking to sell to the national and international market. Relevant data layers were reclassified for each of the three cage types. Figure 2 shows the overall structure of the model. The submodels were based on best available data and constructed following discussions with aquaculture experts. The layers within the submodels are combined using weighted linear Multi-Criteria Evaluation (MCE) (Eastman, 2016). An MCE is a decision support function which F I G U R E 2 Overall model structure for cage aquaculture site selection and zoning of Volta Lake combines a weighted set of criteria to achieve a single composite outcome focussed on a specific objective. Some sub-models had multiple outputs depending on the three cage types. The final output was produced by adding each of the four sub-models together and then multiplying by the Boolean constraints layer. All values above 18 were classed as 'highly suitable', and all values between 17 and 18 were classed as 'suitable' for development of aquaculture in each of the cage size categories.

| Constraints
Certain areas of the lake surface will be unavailable for cage aquaculture due to conflicts with other major activities and environmental factors. The shallowest bathymetry up to 10 m was reclassified as a constraint. The 3 km mandatory buffer zone from Akosombo dam was also classified as a constraint. Landing stages and riparian market centres were identified from Roche (2014), and a 2 km buffer zone was delineated around such centres to avoid conflict with movements as well as potential negative environmental interactions.
Similarly, a 3 km buffer zone was delineated around major ports. A 1 km buffer zone was delineated around existing and planned ferry routes. As the lake is a significant transport corridor, existing and planned major shipping lanes were identified from Roche (2014) for which a 1 km buffer zone was delineated.

| Bathymetry Sub-model
The bathymetry sub-model comprised of a water depth layer. Cages must be situated in locations where there is sufficient depth below the net to allow for water exchange and waste dispersal (Beveridge, 2004). For lakes and reservoirs such as Volta where water level fluctuates, it is important to evaluate potential changes in depth to avoid locating cages in areas that may not be suitable throughout the culture cycle (Ross et al., 2011). As there is no comprehensive depth map for Volta Lake, bathymetric data were generated using an empirical approach based on Landsat 8 images. Light passing through water becomes attenuated, so shallow areas appear bright whereas deep areas appear dark in an image (Gholamalifard et al., 2013;Lyzenga, 1981). As Landsat 8 imagery contains real values reflected from objects in the environment, it is possible to approximate water depth from differential reflectance in the bands. Band 3 (Green, wavelength from 0.525 to 0.600 µm) has the highest penetration in water and was used to calculate water depth, while band 6 (Short Wavelength Infrared) was used to define the water body outline because of its clear reflection from the water surface and minimal atmospheric interference. Approximate water depth was calculated as described by Landmap Geoknowledge (2005) and depends on the empirical tuning noted by Jerlov (1976), Lyzenga (1981) and Jupp (1988). Depth in metres is approximated by: Annual fluctuation in water level in Volta Lake can be up to 10 m due to seasonal rainfall and draw-down for electricity generation.
The satellite data were collected in March 2014 which are near the start of the rainy season, so the bathymetry at this time would be the shallowest of the year, thus allowing modelling of the worst-case scenario as this would be the limiting factor for siting a cage. A number of assumptions were made in the calculation, so the depths can only be indicative of the range found in the lake. However, the calculated depths correlated well with the results of a bathymetric survey of a small section of the lake in Stratum I and with spot measurements taken throughout the lake strata during fieldwork. The final bathymetric layer ( Figure S1) was reclassified based on specifications of small and medium rectangular cages and large circular cages (Table 2).

| Hydrography sub-model
The hydrography sub-model was developed from two key parameters: current velocity and wave height.
Current velocity is an important factor in cage aquaculture as higher current velocity allows good water exchange through fish cages which removes waste from feeding in addition to replenishing oxygen in water (Beveridge, 2004). However, if currents are too strong they can damage cages, reduce net volume and stress fish (Yamazaki et al., 2019). Wave height is also an important consideration for cage aquaculture as high waves can stress or damage cage infrastructure, create unsafe working environments and poor welfare conditions for the fish .

Current velocities were measured quarterly between June 2015
and March 2016 inclusive. Ten 40 × 40 cm drogues, 5 set at 1 m and 5 set at 5 m depths, were deployed at locations within each stratum on the Lake to determine the current velocities ( Figure 1). The initial time of deployment and the GPS coordinates of the location at initial deployment were recorded using a Garmin GPS76, and this was repeated at 30-min intervals for 10 h. The current velocities at the two depths were calculated from the field data and digitized in ESRI ARC GIS version 10.1 to create choropleth raster files depicting current velocities in each stratum.
Current velocity in the lake varied spatially and seasonally and an aggregate of the annual data is shown in Figure S2. An overall, current velocity layer was developed by combining the surface (1 m) and midwater (5 m) velocities in an MCE with weightings of 0.3 and 0.7 respectively. This layer was reclassified to develop a current velocity suitability layer ( Figure S3) using the criteria in Table 3.
Maximum significant wave height was calculated using the approach developed by the US Army Corps of Engineers (1984), which has also been used by Scott (2004) and Falconer et al., (2013) in aquaculture site selection studies This approach uses data on wind direction and speed, fetch and water depth and is suitable for water depths less than 90 m, such as in Volta Lake. Wind data were collected from 4 weather stations around the lake: Akuse TerrSet so that each pixel in the final wave height layer ( Figure S4) had the maximum value of the corresponding positions in the input layers (Eastman, 2016).
Different cage sizes and construction materials have different tolerances to waves which must be considered when selecting sites and cages for aquaculture. Most studies on wave exposure have focussed on cages in exposed areas where the wave height is often over 2 m and the system design criteria and tolerances are well defined (Huang et al., 2008;Ma et al., 2009;Pérez et al., 2003).
However, as the maximum wave heights predicted in Volta Lake were <1 m ( Figure S4), and cage tolerances less defined, wave height classifications were based on expert opinion and experience of cage suitability at different wave heights (Table 4).
The final reclassified current speed and wave height layers were combined in an MCE to create the Hydrography sub-model. Both layers were considered to be of equal importance so were given equal weights.

| Water Quality Sub-model
Good water quality is essential for production, health and welfare of farmed fish and parameters must be within the tolerance ranges for the selected species (Beveridge, 2004), in this case, tilapia. The  The water quality parameters measured were temperature, pH, dissolved oxygen, turbidity, conductivity, total phosphate, nitrate-nitrogen and ammonium-nitrogen, biochemical oxygen demand, faecal coliform and Escherichia coliform. Temperature, pH, dissolved oxygen, turbidity and conductivity were measured insitu with a YSI EXO 2 Sonde. Total phosphorus was determined using the persulphate digestion method. Nitrate-nitrogen and ammonium-nitrogen were determined using the hydrazine reduction and direct Nesslerization methods respectively. All the analyses were according to procedures described in Standard Methods for the Analyses of water and waste water (APHA-AWWA-WEF,

2005).
Suitability classification of the lake water quality for tilapia aquaculture was determined using a modified version of the Canadian

Council of Ministers of Environment (CCME) Water Quality Index
(CCME-WQI) (Equation 2), which is a tool for simplifying the reporting of water quality data (CCME, 2001). The CCME-WQI objectively compares measured water quality values to guidelines for a given activity (in this case criteria for production of tilapia; see Table 5) to produce a score ranging from 0, representing worst quality, to 100, representing the best quality for the given activity.
The CCME-WQI is generated by Equation 2: where F 1 (Scope) represents the percentage of variables that do not meet their objectives at least once during the time period under consideration, relative to the total number of variables measured; F 2 (Frequency) represents the percentage of individual tests that do not meet objectives; F 3 (Amplitude) represents the amount by which failed test values do not meet their objectives. Computation of F 1, F 2 and F 3 are available in detail in (CCME, 2001). The index scores were reclassified based on the protocol in Table 6. The criteria or test values are those defined as within optimal ranges for tilapia culture (Beveridge & McAndrew, 2000).
Some variations in concentrations of selected water quality parameters were observed during monitoring. Consequently, a CCME-WQI was calculated to reflect conditions in the four seasons in northern Ghana; the dry season (January -March) pre-rainy season (April to June), rainy season (July -September) and post-rainy season (October -December). The four seasonal water quality suitability models ( Figure S5) were combined using an MCE with equal weighting to generate an overall combined water quality suitability sub-model.

| Access Sub-model
Physical access to sites is a key factor in the success of a cage site.
Two major factors were considered: population size and transportation costs taking into account distance over roads of differing quality.
Transportation cost from or to a fish farm influences fish price and is affected by good road access and the road type. Access to farm sites for supply of goods and services from cities and small towns is also very important (Ross et al., 2011), as this will bring in labour, feed, equipment, spares and fuel to the site and will also allow access to the fish markets. Similarly, the distance to small towns and The size of population per unit area can be used to approximate the size of markets (Sato et al., 2012). The Ghanaian road map, which identifies 23 road types, was reclassified based on road condition to three categories based on the likely relative cost of transportation: 1 = 'expensive', 2 = 'medium', 3 = 'inexpensive'; footpaths and bridleways were ignored.
The access sub-model was developed using the TerrSet COST process, based on transportation cost along roads and over water starting from the 170 districts, and also the populations in each district. According to Taiwo and Kumi (2013), the average transportation cost for a 'good quality' road is Ghȼ 0.90 per tonne-kilometre to the market centres, Ghȼ 1.25 per tonne-kilometre for 'medium quality' roads and Ghȼ 1.43per tonne-kilometre for 'poor quality' roads.
Road and lake cost layers were combined to form a single cost layer from which the cost of product transportation from any lake location to any district centre was calculated. Each district cost map was divided by the head of population in the district to produce the cost of transportation from each district per person. This layer was then reclassified using the values in Table 7 to produce two access sub-models: small cages and medium/large cages. Farmers operating small cages were assumed to have limited access to capital to invest for transportation and other equipment, so proximity to potential local markets would be more important. Medium and large cage operators are more likely to deal with larger national or export markets, so proximity to roads or railways would be much more advantageous.

| Constraints
The constraints layer indicates areas that would not be available for aquaculture development (Figure 3). All parts of the lake have areas where constraints prevent development of cage aquaculture. The major constraints are the transport routes located throughout the entirety of Volta Lake. The North West of the lake has a large area, in strata VII and VIII, that would be unavailable for aquaculture due to concentration of transportation routes.

| Bathymetry sub-model
Highly suitable and suitable areas of Volta Lake available for cage aquaculture based on water depth are mostly located in the southern and central parts of the lake (Figure 4). Highly suitable areas for the small cages are in the margins of the lake, although much of the area is classed as suitable with major exceptions being the deeper parts of the southern section. Almost all of the lake has either highly suitable or suitable depths for medium cages with low scores in the shallow northwest and in the deepest parts of the southern basin. Much of the southern basin and central section of the lake has highly suitable depths for larger cages.

| Hydrography sub-model
The results of the hydrography sub-model suggest there are a greater number of suitable locations for small and large cages than medium size cages ( Figure 5). Suitable locations for small cages are found in the north, west and mid-sections of the lake. While for the large cages, suitable locations are found in the mid-sections and south of the lake. This is a reflection of the water exchange in these areas as the higher volume of fish in the larger cages require consistently faster water flows, which are found in wider areas in the mid and south areas due to wind effects on the water surface, and being nearer the outflow of the lake in the south.

Dry season (Jan-March)
Pre-rainy season (April-June) TA B L E 7 Seasonal variation in CCME-WQI water quality index Volta Lake

| Water quality sub-model
Based on the CCME-WQI, water quality in the different strata, as for the hydrography, varied seasonally and spatially (Table 8). The reclassified suitability of the lake water quality ranged from suitable to moderately suitable for cage aquaculture in the southern parts and unsuitable in the northern section ( Figure 6). The most southerly and central sections of the lake had the overall highest water quality, whereas the Afram arm (stratum I) and the northerly sections were least suitable.
Overall, 54.2% of the lake area was classed as moderately suitable, the remainder being unsuitable (low for cage aquaculture) ( Figure 6).

| Access sub-model
The sub-model outcome (Figure 7) indicates that for transport costs, the highly suitable and suitable areas for small cages are located around the edge of Volta Lake where small scale fish farmers would have easy access to facilities and to markets. Over 90% of the lake area is highly suitable or suitable for larger scale commercial scenarios with only small areas in the centre and northwest of the lake being unsuitable. In comparing these two scenarios, there are greater opportunities for medium and large cages as highly commercial aquaculture will deploy better cage technology, management systems, transport facilities and have better ability to serve more distant markets. Although the products from small cage farms are less easily moved over large distances, they are nevertheless important as they are traded locally and consumed by local families around villages.

| Final model outputs
The most suitable areas for cage aquaculture vary depending upon cage size and technology level (Figure 8). When constraints are applied, much of the northern Black Volta (strata VII and VIII) and Oti (stratum V) arms of the lake are unsuitable, whereas substantial suitable areas for cage aquaculture are located in the main body of the lake. A greater proportion of the lake is classed as suitable or highly suitable for medium and large cages in comparison to small cages (Table 9). The most suitable areas for small cages are near the lake shores, medium sized cages can occupy more of the lake area while large cages can take advantage of much of the lake surface scoring highly especially in the south. This is illustrated well in the southern parts of the lake (see expanded images) where there is a considerable area that is highly suitable for large cages, a more limited area for medium and small cages.

| DISCUSS ION
The study used GIS to outline potential suitable locations for cage aquaculture in Volta Lake. The modelling approach used data from a range of sources to assess whether areas throughout the lake met the specified requirements for cage culture of tilapia. This is a similar method to other aquaculture studies that have used GIS for site suitability assessment and site selection studies (Aguilar-Manjarrez & Ross, 1995;Falconer et al., 2013;Longdill et al., 2008;Pérez et al., 2005;Ross et al., 2011;Salam et al., 2003). Similar to Ross et al., (2011) andFalconer et al., (2013), several different cage sizes were modelled to support decision makers in identifying the most suitable areas for a range of cage systems. Differences in suitable areas were apparent in the sub-models and in the final model output, highlighting the importance of modelling multiple cage dimensions when evaluating suitability of large lakes for aquaculture. The depth layer gives a good approximation of depth based on remote sensing data; however, there are substantial areas of the lake with submerged trees which are an impediment to cage installation. As no specific data were available, these areas were not included as constraints but would need to be taken into account when installing cages.
Availability and quality of data is a major constraint for any spatial assessment (Falconer et al., 2020). Although Volta Lake has been studied previously for different purposes, it is still a data scarce location and there is a lack of high-resolution lake-wide data sets that capture the spatio-temporal heterogeneity of parameters of importance F I G U R E 4 Bathymetry suitability sub-model for small, medium and large cages in Volta Lake (above) and in enlarged southern parts of the lake (below) to aquaculture. This study has used data from a number of different sources, but there are still limitations. For hydrography and water quality measurements, sample stations were used to characterize the stratum of lake and it is recognized that variations in the area will not be captured. Selection of station location was based on previous experience and knowledge of conditions in the lake. Additional sample sites and more frequent sampling would improve the underlying data and, in turn, would improve the spatial models, but the results nevertheless provide a general overview of the lake conditions. This then allows future effort to be targeted on the highest scoring areas as these areas are most likely to be suitable for aquaculture.
The model outputs identify the most suitable locations based on the input criteria and are used to identify areas for more detailed analysis and fieldwork. The number of cages or biomass of fish that can or should be established within the lake depends on other factors including regulations and carrying capacity .
For large, complex, lake systems such as Volta that have many branches and varying hydrological conditions, local scale assessment is particularly important. With more location specific data, models such as Dillon and Rigler (1975) and OECD (1982) can be used to calculate production and ecological carrying capacity and determine the level of fish production that could take place at individual sites. Furthermore, land use in the surrounding catchment and nutrient loading into the lake will also influence local carrying capacity and production potential (Milne et al., 2015;Paterson et al., 2006). Further research is required to develop improved modelling F I G U R E 5 Hydrography suitability sub-model for small, medium and large cages Volta Lake (above) and in enlarged southern parts of the lake (below) Regulatory authorities, government bodies or coordinating organizations can develop and implement zones, but these should also be supported by relevant policies and legislation to ensure aquaculture is planned and managed appropriately. Not all areas designated as suitable will necessarily be incorporated into aquaculture F I G U R E 7 Access sub-model for two different farm size scenarios in Volta Lake (above) and in the southern parts of the lake (below) zones. This is especially important if aquaculture is to contribute to meeting the SDGs as shared resources such as freshwater lake systems are often under pressure from many different user groups and a coordinated approach to optimize use of lake resources for all activities is necessary. The outcome of this study has significant potential to contribute to implementation of the newly instituted

| CON CLUS IONS
The great lakes of the world, and the growing number of very large reservoirs such as Volta Lake, provide a range of ecosystem services including hydropower, transportation routes, water for domestic and agricultural purposes as well as fish from principally artisanal fisheries. They are also prime sites for production of fish in floating cages, which can contribute to national economies and to international food security. Any expansion of the cage aquaculture industry must be contained within the lakes' capacity to assimilate wastes without compromising the supply of ecosystem services or the access and benefit rights of riparian communities.
A first step in sustainable development of aquaculture is to identify potential suitable locations. Broad-scale, lake-wide, assessment using GIS modelling is a cost-effective approach that allows resources to be targeted in the most appropriate locations to support sustainable management of cage aquaculture in Lake Volta, Ghana, and elsewhere. The approach developed for three cage sizes can provide objective support for decision-making, minimizing conflict with other water users. In addition, based on transportation cost and population distribution, we have shown the spatial distribution of the important effect of market access to suitability for aquaculture development. This combination of sub-models can be used together with more local scale models of environmental carrying capacity and environmental quality to determine prospective sustainable yield from aquaculture in the large water bodies of the world, and as a starting point for development of aquaculture management zones.

ACK N OWLED G EM ENTS
The study was undertaken as part of the project 'Planning for Improved and Sustainable Aquaculture in Volta Lake, Ghana', which was funded by a Leverhulme-Royal Society Africa Award Institute, Accra, who provided assistance with GIS.

CO N FLI C T O F I NTE R E S T
The authors affirm that there are no sources of conflict of interest pertaining to this work.

DATA AVA I L A B I L I T Y S TAT E M E N T
The data that support the findings of this study are available from the corresponding author upon reasonable request.