Lake Victoria supports the largest inland freshwater fishery on earth. The Kenyan portion of the lake covers an area of 4128 km² and produces the bulk (80%) of the country’s annual fish landings. The lake had high fish species diversity of over 500 species of fish most of which were endemic but following the introduction of Nile perch and tilapiine cichlids, coupled with over exploitation and environmental degradation, many native species have since disappeared from the lake. The main commercial species caught are Rastreonobola agentea (Omena/dagaa/mukene) 53.32%, Lates niloticus (Nile Perch) 33.4% and Oreochromis niloticus (Tilapia) 4.31%. Fish stocks have dwindled due to heavy eutrophication, proliferation of invasive species, over-fishing and use of illegal/undersize gears.

Ecological processes at various temporal and spatial scales have created a vast array of distribution patterns (Nyamweya et al., 2016). The configuration of habitats and the distribution of species along environmental gradients have for a long time also been affected by human interests and development. For example, in 1877, applied ecology sought the answer to an unexpected low production of an oyster bank in response to fisheries management needs. Today, as habitat loss poses a major threat to ecosystems (Aryaguna, 2019), the need to map species and habitat distributions for nature conservation has increased dramatically (Vayghan et al., 2013). Ecologists’ interests in understanding population distribution and dynamics, combined with advancements in computer technology, have led to the development of species distribution modelling. This field of spatial modelling relates the occurrence or abundance of species to environmental descriptors and can help resolve, or raise, ecological and evolutionary questions into the underlying processes and mechanisms regulating species and habitat distributions.

The objectives of this study were to:

1) To identify the habitat variables that promote fish breeding.

2) To extract lake surface temperature and chlorophyll-a from satellite imagery.

3) Determine potential fishing grounds using Habitat Suitability Index (SI).

Lake Victoria traverses the three East African Countries: Kenya, Uganda and Tanzania. It lies within longitudes 33˚52'55'' and 35˚00'45'' East and latitudes 0˚22'42'' North and 1˚00'34'' South. It has a surface area of 68,500 km² shared between Kenya (6%), Uganda (43%) and Tanzania (51%). The Kenyan portion of the lake covers an area of 4128 km².

The climate in the lake basin varies from tropical rain forest with rainfall over the lake for much of the year to a semi dry climate with intermittent droughts over some areas. Figure 1 shows the study area.

Landsat images for the months January, May and December, 2019 were acquired from USGS website were used. Fishery data was acquired from Kenya Marine and Fisheries Research Institute (KMFRI) where different fishing methods was represented in the data. The fishery data together with the suitability index was used to validate the remote sensing images. Table 1 gives an indication of the data used in this research.

For this research, the satellite images were imported, reprojected to change the

projection to UTM. Subset was extracted to obtain area of interest or region of interest (subset image). Image correction such as radiometric calibration, atmospheric correction for cloud filtering, then lake surface temperature and chlorophyll-a were extracted from the image (see Figure 2).

The estimated data was obtained by the Landsat 8 Thermal Infrared Sensor (TIRS). Landsat 8 provides metadata of the bands such as thermal constant, rescaling factor value (Figure 3).

First, the digital number (DN) of band 10 and band 11 had to be converted to radiance by using following Formula (1);

L λ = M λ × Q c a l + A λ (1)

where:

L λ is the top of the atmosphere spectral radiance,

M λ is band specific multiplicative rescaling factor,

Q c a l is digital number,

A λ is the band specific additive rescaling factor.

Then, converted them to the brightness temperature by using this Formula (2);

T = K 2 ln ( K 1 L λ + 1 ) (2)

where T is at satellite brightness temperature (Kelvin), K₁ and K₂ are thermal conversion constants from metadata.

Calculation of NDVI was important because subsequently the proportion vegetation (PV), which is highly related to NDVI and emissivity(e) which is related to PV must be calculated using Formula (3).

NDVI = Band 5 − Band 4 Band 5 + Band 4 (3)

Proportion of vegetation was computed to estimate the emissivity. Usually, the maximum and the minimum values of NDVI image are displayed directly in the image in ArcGIS, therefore PV is calculated as follows in Formula (4);

PV = ( NDVI − NDVI min NDVI max − NDVI min ) 2 (4)

Emissivity was computed as follows using Formula (5):

e = 0.004 ∗ PV + 0.986 (5)

in the raster calculator.

The value 0.986 correspond to correction value of the equation and 0.004 is a constant as illustrated by USGS website.

Finally, the LST equation was applied to obtain the surface temperature map using Formula (6).

LST = BT 1 + ( λ ∗ BT / C 2 ) ∗ ln ( e ) (6)

C2 = 14,388 µm

The values of λ for Landsat bands include.

Chlorophyll-a was extracted as follows, since the Landsat-8 data (Table 2) was stored in digital number (DN). It had to be radiometrically converted to the top-of-atmosphere radiance (L_TOA) by using following Formula (7);

L λ = M λ × Q c a l + A λ (7)

where:

L λ = TOA spectral radiance

M λ = Band specific multiplicative rescaling factor

Q c a l = Digital Number

A λ = Band specific additive rescaling factor

After obtaining the radiance value, the next step was atmospheric correction that will automatically convert the top-of-atmosphere radiance value (L_TOA) to bottom of atmosphere reflectance using FLAASH in ENVI 5.3. FLAASH incorporates the MODTRAN radiation transfer code. It enables corrections for the adjacency effect or pixel mixing due to scattering of surface-reflected radiance.

The chlorophyll-a was extracted using band 2 and band 4 in Landsat image as follows (Formula (8));

Chl_a = Band4 / Band2 (8)

The above processes were repeated for all the images (January, May and December 2019).

Fish catch frequency were analyzed to determine the preferred range of the fish by relating the availability of fish catch data with the satellite images (Wanchana & Sayan, 2018). Fishing effort is often considered to be an index of fish availability or fishing occurrence. This is to derive suitability index. Habitat suitability index (HSI) models are widely used as a tool in fisheries management, ecological impact assessments and ecological restoration studies (Tidd et al., 2018). The HSI models usually describe relations between fish abundance and ecological variables and estimate the level of habitat suitability. HSI objectively assess the range of environmental condition that fully, marginally or do not meet the requirements of a species. Output produced from HSI modeling can predict the spatio-temporal variation of fish habitat conditions and can be used in combination with GIS to provide maps and information upon which managers can make informed decisions in fisheries management (Mustapha et al., 2010).

Suitability Index Formula (9);

SI = Y f i t − min Y f i t max Y f i t − min Y f i t (9)

where Y f i t is the predicted value of frequency (CPUE); min Y f i t and max Y f i t are, respectively, the minimum and maximum frequency (CPUE). The suitability index determined the level of scores of the preferred ranges for fish. The satellite images were reclassified using these relevant scores to derive map of potential fishing area.

The lake surface temperature was obtained for the months January, May and December. The red regions represented high temperature areas and green represents low temperature areas. For January the high temperature was observed along the lake shores of Kisumu, and parts of Siaya and Busia counties. Optimum temperature for fish breeding were along Homa bay county and some along Siaya counties (Figure 4). Trisakti et al. (2004) analysed sea surface temperature using Landsate 7 ETM to understand potential fishin zones. Syariz et al. (2015) also did the same process for ecological studies.

In May high temperature was along Kisumu and Homabay while other counties the temperature were varying and cooler were in the interior of the lake (Figure 5).

December recorded higher temperatures along Kisumu and Busia counties and lower along other counties and parts of Mbita where the temperature was optimum (Figure 6).

High chlorophyll-a concentration areas i.e., along shores of Kisumu, Homa bay along Rangwe, Mbita, Karachuonyo and some parts along Siaya county. This is an indication of thriving phytoplankton which fish feed on. The dark blue regions are those which show high chlorophyll-a concentration while the reddish regions show low concentration. The optimum chlorophyll-a concentration was observed in the Homa bay and Siaya county beaches. This was constantly observed throughout the three months; January, May and December 2019 (Figure 7(a)-(c)). Such reports have been dervived by Apriliani et al. (2018) in their studies. Similar studies were done by Lailia et al. (2015).

From this study the distributions of the essential fish habitats were observed along beaches of Homa bay, Mbita region, this is due to abundance of chlorophyll-a in these regions as observed above. From Figure 11 below the distribution is represented in terms of the suitability index obtained from the CPUE. The low habitable regions are regions which registered catch of less than 10

metric tonnes, and were observed around Kisumu County. Highly habitable regions were observed along the Mbita in Homa bay county and some in Siaya county, they registered catch greater than 30 metric tonnes. Other regions recorded between 10 and 30 metric tonnes as observed from the fishery data.