Abstract:

This study aims to estimate crab yield by using meteorological factors. The crab yield and meteorological factors were acquired at five key growth stages (stocking period, 1st-3rd shelling period, 4th shelling period, 5th shelling period and mature fishing period) from 2010-2022 in Xinghua of Jiangsu. Meanwhile, the meteorological factors were used as the inputs for the three machine learning algorithms for the crab yield estimation. The three machine learning algorithms included the Random Forest Regression (RFR), Support Vector Regression (SVR), and Partial Least Squares Regression (PLSR). Finally, the estimation models were developed using Leave-One-Out cross validation under different growth stages. The results showed that the precision of estimating crab yield by different meteorological factors was significantly different at different growth stages. For the three machine learning algorithms, the SVR was found to be superior to the RFR and the PLSR (coefficient of determination, R²=0.95-0.98, RMSE=24.07-35.14 kg/hm²). In the single growth stage, the highest estimation accuracy was observed at the stocking period (R²=0.98, RMSE=24.07 kg/hm²). In the whole growth period, the estimation accuracy was (R²=0.92, RMSE=44.31 kg/hm²). Therefore, the meteorological factors can be expected to estimate crab yield under the different growth stages.

Key words: crab yield, meteorological factors, RFR, SVR, PLSR