基于估计准度和调查费用的渔业资源调查采样设计优化

合理的调查采样设计及优化可以降低调查费用,确保调查数据的准确性,实现调查采样效益的最大化。本研究量化了调查费用,在考虑估计准确度基础上,将调查费用作为评价指标加入到采样设计优化中,应用计算机模拟重抽样方法和模拟退火算法进行航线规划。基于2013—2015年春、秋季海州湾渔业资源底拖网调查数据,使用克里金插值法模拟了该海域星康吉鳗(Conger myriaster)、方氏云鳚(Enedrias fangi)和大泷六线鱼(Hexagrammos otakii)在春、秋季的相对资源量分布,作为其分布的“真值”。应用分层随机抽样模拟不同样本量下的调查采样,估计各目标鱼种的资源量指数,以相对偏差(RB)的绝对值来评价目标鱼种资源量指数估计的准确度,以调查费用、超标概率作为费用评价指标,以综合评价指标(IEI)作为衡量准确度和费用的综合指标。结果表明: 各鱼种资源量指数估计的RB值随样本量增加而降低,但各鱼种的RB值不同。不同评价指标确定的最优站位数不同,按RB值确定的最优站位数较高,调查费用超过预算;按调查费用确定的站位数较低,但无法有效保证资源量指数估计的准确性;按IEI确定的最优站位数处于中间水平,能兼顾调查费用和资源量指数估计准度。综合考虑调查费用和航线规划的采样设计不仅能确定最优站位数,还可以确定调查设计的站位空间位置、站位调查顺序和调查费用。调查者可以根据实际费用预算,从模拟方案中选择费用低、RB低的理想采样设计方案开展调查。

Sampling design optimization based on estimation accuracy and survey cost for fishery-independent surveys

Optimization of sampling design can reduce survey cost, ensure the accuracy of survey data, and get the maximum benefit of survey design. In this study, survey voyage was added into sampling survey design optimization as the evaluation index to quantify the survey cost. Computer simulation and resampling technique were used to simulate the survey plan. Simulation annealing algorithm was used to find the survey design with the shortest voyage. Based on the survey data collected from the bottom trawl survey conducted in the Haizhou Bay and its adjacent waters in spring and autumn from 2013 to 2015, Kriging interpolation was used to simulate the relative abundance distribution of Conger myriaster, Enedrias fangi, and Hexagrammos otakii in the bay in two seasons as their ‘true’ values. Resampling was conducted using stratified random sampling with different sample sizes for simulation study, and the abundance indices of each target species were estimated based on the simulated data. The relative bias (RB) was used to evaluate the accuracy of estimation of abundance index. The average survey cost and probability of budget overshoot (P) were used to measure the cost of survey. Integrated evaluation index (IEI) was developed to measure the survey cost and estimation accuracy comprehensively. The results showed RB values of all target species decreased with sample size. Different target species had different RB values with the same sample size. The optimal numbers of station determined by different evaluation indices were different. The optimal sample size determined by RB was relatively high, while the cost exceeded the budget of survey. The optimal sample size determined by probability of exceeding survey budget was relatively low but the accuracy and precision of estimation was low. IEI balanced the survey cost and estimation accuracy, and the optimal sample size defined by IEI was at intermediate level. The sampling design considering survey cost and route planning could not only determine the optimal sample size, but also record the potential station location, the survey sequence and the corresponding survey cost of simulations. According to the actual cost budget, the desired survey design with low cost and low RB of estimation from simulations could be chosen for fishery-independent surveys.