3. MATERIALS AND METHODS

3.1 Data: origin and treatment

Fish yield data collection in both Zambia and Zimbabwe has been in operation since the fishery started in 1961 (Zambia) and 1962 (Zimbabwe). In addition both the Lake Kariba Research Unit (LKRU) in Sinazongwe (Zambia) and the Lake Kariba Fisheries Research Institute (LKFRI) in Kariba town (Zimbabwe) (Figure 1) have carried out extensive monitoring with experimental gillnets. Very little data, however, are available from Zambia in the period 1974-1980 when the fishery was officially closed due to the Zimbabwean civil war.

Hydrological data were obtained from the Zambezi River Authority (ZRA) as daily lake level recordings, which were averaged on a monthly and annual basis (Figure 2).

3.2 Inshore fishery, Zimbabwe

Fisheries catch and effort data collection has been in operation since the fishery started in 1962 on the Zimbabwean side. Until 1992 a data collection system referred to as Catch Assessment System (CAS) was in place. From 1993 the CEDRS system, unified with Zambia, was introduced as one of the activities under the Zambian-Zimbawean SADC Fisheries Project. Artisanal fishing on the Zimbabwean shores of the lake is conducted in seven fishing areas (C1 to C7, Figure 1). A description of these fishing areas and changes that they underwent during the history of the fishery is given in Marshall, Junor and Langerman (1982) and Karenge and Games (1995). A compilation and partial analysis of the catch-effort data on the Lake Kariba artisanal fishery in Zimbabwe is given in Songore (2000). Methods of data collection differed according to how fishing activities were organized. Operators of concessions (fishing areas C1, C3 and part of C5) were required to submit monthly returns of their daily catches, effort, and sales. The annual catch of this group was worked out from the returns. Data from the remaining fishing areas operated by the local communities were collected through an enumeration programme. Staff from LKFRI carried out enumeration in two fishing areas (C2 and C4) while staff from the Binga District Administrator’s office enumerated catches from the three other areas (part of C5, C6 and C7).

Prior to the intensification of the liberation war in Zimbabwe, data were collected from generally all the fishing areas. Later, during the period 1975-1979, when the war escalated, data collection operations were disrupted with some of the sampling areas being abandoned due to closure of fishing villages. This mostly affected data collection in fishing areas that were fished by local communities. The concessionaires continued sending in their catch, effort and sales returns undisrupted. Attempts to normalize data collection again were made in 1980 when the war ended. However, due to inadequate manpower and financial constraints, it was not possible for LKFRI to cover all the landing sites using its own resources. Hence the Binga District continued for some time to collect data from landing sites under their jurisdiction. But, data from these areas (C5 to C7) were sent in irregularly. From 1973 to 1984 total yield was estimated from mean monthly catch from all villages (where catches from villages not enumerated were estimated) through multiplication by 12 months and adding the reported catches from the concessionaires. From 1985 some of the concessionaires were transformed into cooperatives, and during the period 1985-1992 enumeration activities were scaled down to ten fishing villages which were all in the areas C2 and C4. Therefore yield estimates for the whole lake were based on extrapolation of data collected from C2 and C4.

A critical analysis of the sampling strategy and a number of recommendations were made by Sanyanga, Lupikisha and Thorsteinson (1991). This led to the introduction in 1993 of the CEDRS in order to unify the data collection systems of Zambia and Zimbabwe. As it was found impossible to sample either all villages or the total landings for any particular village for the whole year, ten representative villages were to be enumerated for ten days every month on each side of the lake. The sampled catch was to be raised to an estimate of the total catch by multiplying by the ratio between total number of days sampled and days in the year and the ratio between the number of fishers in the villages sampled and the total number of fishers. CEDRS only operated well for a short period, but broke down due to diminishing human resources. Enumerators who retired, resigned or died were never replaced. Furthermore, very few of the proposed strategies for improving were implemented. Hence the quality of data collected continued to be questionable in terms of representing the whole fishery (Songore, 2000).

Total yield estimates from 1985 to present have been calculated using the following formula:

⁽¹⁾ total # days = 365 until 1993, thereafter = 360.

⁽²⁾ from 1985 to 1992 total # fishers used in the calculation was constant (= 765), from 1993 the total # fishers was obtained from Frame surveys, keeping the number constant in between each survey.

3.3 Inshore fishery, Zambia

Scholz (1993) gives a detailed account of various sampling designs used and an analysis of the data from 1961-1990. Aggregated data from his thesis are used for the period 1961-1980. Data from 1980 to 1999 are based on a compilation and partial analysis of the catch-effort data on the Lake Kariba artisanal fishery in Zambia by Musando (2000). Two different systems of collecting catch-effort data have been used. From 1980 to 1993 the so-called CAS (Catch Assessment Surveys) was used, a two stage stratified random sampling system to estimate catch rates (Cpue) supplemented by a framesurvey to count total effort (f) and activity level (A), based on which total catch (C) could be estimated (). During three survey days at randomly selected sampling units (PSUs), fishers met at their landing sites were asked to report on catch, mesh sizes and number of nets set. Settings with no catch are said to have been reported as well. The catch was identified by fish family and weighed in kilos. The “CEDRS” (Catch and Effort Data Recording System) was introduced during 1993; in order to unify the data recording system with the Zimbabwean side of the lake. The CEDRS programme was designed to carry out the survey for ten days per month per stratum (Figure 1), in ten selected fishing villages, as well as record catch rate by species name other than the family name.

Unfortunately the execution of the sampling programme for both the CAS/CEDRS designs has been highly irregular due to budget shortcomings and lack of staff (Scholz, 1993, Musando, 2000). Moreover, it appears that the total estimated effort is made as a simple summation of strata sampled, irrespective on the number of strata actually sampled. In this way, the “effort” thus becomes a function of number of strata and on the number of times each strata has been sampled during a year (Musando, 2000).

3.4 Frame surveys on Lake Kariba

Lake wide frame surveys on Lake Kariba were initiated in 1990 under the auspices of the Zambia/Zimbabwe SADC Fisheries Project (ZZSFP) and repeated in 1993. Additional frame surveys have been conducted in 1995 and 1999 on the Zambian side and in 1998 and 2000 on the Zimbabwean side.

3.5 Experimental data

Since 1960 LKFRI in Zimbabwe has conducted an experimental gillnet fishery at Lakeside station close to Kariba town in the Sanyati basin (Basin 5, Figure 1). Here commercial fishing was only permitted between 1962 and 1972 (Kenmuir, 1984) and limited fishing took place from 1977 to 1980, but this probably had little effect on the fish stocks (Marshall, 1984). Experimental fishing during the early years was somewhat sporadic with few settings (Kenmuir, 1984). However, from 1970 to 1975 settings were done twice a month and since 1976 sampling has been done weekly, except for 1981 when no fishing took place (Karenge, 1992; Karenge and Kolding, 1995a). The sampling design has remained constant over the whole period and consists of multifilament nets in multimesh fleets with 12 panels ranging from 38 mm to 178 mm stretched mesh sizes with 12.5 mm increments. The fleet, 550 m in total length, was set overnight perpendicular to the shore with the smaller meshes inshore. Each individual fish caught was recorded for species, mesh size, weight, length, sex and maturity stage. All experimental gillnet data from 1969 to 1999 have been digitized into PASGEAR (Kolding, 2001) which now contains more than 185 000 individual fish records from 39 species and represents more than 14 000 panel settings. A detailed analysis of this data set up to 1992 is given in Karenge (1992) and Karenge and Kolding (1995a).

In Zambia, the Lake Kariba Research Unit (LKRU) under DoF, based in the former Fishery Training Centre in Sinazongwe, has carried out an experimental fishing with multimesh gillnets since 1980. Under this programme, every month three experimental localities (stations) were sampled for three consecutive days. Multifilament nets were in use during 1980 to May 1984 and also from 1992 to 1995, whereas monofilament nets were used between June 1984 to part of 1992 and 1993. Both gear types were set at a hanging ratio of 0.5. A fleet of nets with stretched mesh sizes ranging from 25 mm to 178 mm with 12.5 mm increments, were set. A detailed description and analysis of these data up to 1995 is given in Musando (1996). All experimental gillnet data from 1980 to 1999 have been digitized into PASGEAR. The database now contains more than 118 000 individual fish records from 36 species and represents more than 13 000 panel settings. Unfortunately, the choice in stations sampled during this programme has shifted repeatedly, the gillnet fleets were not always complete due to difficulties in replacing worn out nets while the sampling programme occasionally was not carried out because of fuel-shortages, net thefts and bad weather. For the present analysis, due to the irregular sampling design, only the most used sampling stations in close proximity to Sinazongwe (Charlets Island (st.12), Samaria Island (st.21), Zongwe estuary (st.42), Ngoma (st.82), and Nang’ombe (st.132)), have been included (Figure 1). From 1994 to 1995 only Charlets Island and Zongwe estuary were sampled. From 1996 to 1999 only Zongwe estuary remained in the sampling programme. Furthermore, only the mesh sizes in the range 50 mm stretched to 152 mm stretched have been included as only this part of the fleet was more or less complete (except for 1989 when only mesh sizes from 89 mm to 152 mm were used).

Catch per unit effort (C/f) in the both the Zambian and Zimbabwean experimental fishery is expressed as catch (kilo wet weight or number of fish) per standard gear unit per setting. The standard gear unit of each mesh size is one panel of 45 m (50 yards) long.

3.6 Multiple regression of hydrology and fishing effort on catch rates

Multiple regressions were performed with mean annual total catch rates (CPUE) as dependent variable, and hydrological indices and effort as independent variables. From the monthly or annual mean lake levels five differently time-lagged indices for the hydrological dynamics in Lake Kariba were tested following (Karenge and Kolding, 1995b):

The absolute mean annual lake levels (m a.s.l.).

The annual amplitude (m), i.e. the difference between the lowest and highest recorded level within one year.

Delta (D) lake levels (m), defined as the difference between the mean annual lake levels of one year and the immediately preceding year. This will permit the estimation of the effect in change of lake level of one year on the catch rate the succeeding year.

A combination of 2) and 3) by adding the two figures (D lake level + amplitude), thus taking into account the combined effect of two consecutive years.

The D_y + D_y-1 (m), which is the sum of the delta lake levels of one and the preceding year, thus giving the combined effect of three consecutive years.

Fishing effort was expressed as total annual number of nets or fishers (inshore fisheries) or total annual boat nights (Kapenta fishery).

Using cross-correlations, where all variables were made orthogonal by subtracting the mean from the original series, it was established whether there was a lag-phase between detrended annual mean water levels and detrended annual mean catch rates, and which of the hydrological variables was more informative. The rise in lake level normally terminates in May-June every year, whereas the lowest lake levels normally are found in late December-early January. The annual fishery statistics are the average of the whole year, but with generally higher catches in the latter part of the year during low lake levels (Marshall, 1988; Karenge, 1992).

The multiple regressions were always of the form:

CPUE_i = µ_i + effort_i + hydrological index_lag(x) + effort · hydrological index_lag(x) + e_i

In all cases non-significant explanatory variables were removed from the model using stepwise regression. The interaction effect is interpreted as reflecting possible changes in catch rate either as a result of changes in efficiency or of usage of gears in relation to water levels. In case of confounded effects between effort and hydrological variables (i.e. when both variables were significant alone, but one or the other fell out in the combined model) interaction was always high and cannot be interpreted. All statistical analyses were carried out using Statistica (StatSoft ®).