TL;DR
This study investigates the distribution patterns of the economically important Collichthys lucidus fish species in Zhejiang's offshore waters using a random forest model. It reveals that water depth, bottom temperature, salinity, and dissolved oxygen are key drivers of its distribution, with significant seasonal variations, providing crucial data for fisheries management.
What Problem Does This Paper Address?
The East China Sea, particularly the waters off Zhejiang Province, is a vital region for marine fisheries. However, this ecosystem is increasingly impacted by global climate change and human activities, leading to shifts in marine environments and, consequently, fish distribution. Collichthys lucidus, a small but economically significant fish, is a key species in this region. Understanding its spatial and temporal distribution patterns and the environmental factors influencing them is crucial for effective scientific management, conservation, and sustainable utilization of this fishery resource. Previous studies have been limited in scope or outdated, necessitating a more comprehensive and up-to-date analysis.
Methods
The researchers employed a Random Forest (RF) model, a powerful machine learning technique for classification and regression. RF models are known for their ability to handle complex, non-linear relationships and reduce the impact of outliers by building multiple decision trees and combining their predictions. This approach was chosen over traditional methods like generalized additive models (GAMs) because it can better capture the intricate interactions between multiple environmental variables and fish distribution.
The study involved:
- Data Collection: Fisheries survey data on C. lucidus from bottom trawl surveys conducted in Zhejiang's offshore waters during spring (April) and autumn (November) from 2018 to 2022.
- Environmental Data: Collection of physical and chemical oceanographic data, including water depth, bottom temperature (BST), bottom salinity (BSS), bottom dissolved oxygen (BDO), bottom chlorophyll a (BChla), and bottom pH (BpH).
- Data Processing: Calculation of "tail density" (number of individuals per unit area) as a measure of resource density. A multicollinearity test (Variance Inflation Factor - VIF) was performed to ensure that the selected environmental variables were independent enough for the model.
- Model Training and Validation: A portion of the data (75%) was used for training the RF model, and the remaining 25% for verification. Five-fold cross-validation was used to assess the model's prediction performance and accuracy, repeated 100 times for robustness.
- Variable Importance Analysis: The RF model was used to rank the importance of each environmental variable in predicting C. lucidus distribution.
- Relationship Analysis: The study visualized the relationships between tail density and key environmental variables using locally weighted regression (LWR) to understand non-linear trends.
Datasets
The primary dataset used in this study comprises:
- Fisheries Survey Data: Bottom trawl survey data of Collichthys lucidus collected from 120 stations in the offshore waters of Zhejiang Province, China, during April (spring) and November (autumn) from 2018 to 2022. This dataset includes information on the number of individuals caught, allowing for the calculation of tail density.
- Environmental Data: Corresponding measurements of water depth, bottom temperature, bottom salinity, bottom dissolved oxygen, bottom chlorophyll a, and bottom pH collected synchronously with the fisheries data.
This study does not introduce a new benchmark dataset but rather utilizes and analyzes existing survey data over a five-year period, providing a valuable temporal dimension to the analysis.
Key Insights and Ideas
- Seasonal Dominance: C. lucidus exhibits a clear seasonal pattern, with significantly higher densities observed in autumn compared to spring. This suggests a strong influence of seasonal environmental changes on the species' abundance.
- Depth as a Primary Driver: Water depth emerged as the most critical factor influencing C. lucidus distribution in both spring and autumn. This highlights the species' preference for specific depth zones, likely related to feeding, spawning, or predator avoidance.
- Variable Importance Shifts: While water depth is consistently important, the influence of other factors like bottom temperature, salinity, and dissolved oxygen varies between seasons, indicating a dynamic response of the fish to its environment.
- Limited Impact of Chlorophyll and pH: Interestingly, chlorophyll a concentration and pH were found to have no significant effect on the species' distribution, suggesting that other factors are more critical for its habitat selection.
- Spatial Segregation: The study hints at potential north-south population differentiation within Zhejiang's waters, possibly linked to regional climate and oceanographic differences, which could explain spatial distribution variations.
Key Results
- Seasonal Density Difference: The tail density of C. lucidus in autumn was approximately 3.12 times higher than in spring.
- Optimal Habitat Conditions:
- Spring: Peak resources were found in depths of 4-30 m, salinity of 27-34, temperature of 13-16 °C, and dissolved oxygen of 5-9 mg/L.
- Autumn: Peak resources were observed in depths of 4-45 m, salinity of 13-34, temperature of 17-21 °C, and dissolved oxygen of 6-9 mg/L.
- Model Performance: The RF model demonstrated good fitting capabilities, with R² values of 0.873 for spring and 0.912 for autumn. The model's interpretation rates were 23.26% for spring and 34.88% for autumn.
- Spatial Distribution: In spring, C. lucidus was primarily found in northern and central Zhejiang waters, mainly within 31 m depth. In autumn, its distribution expanded to northern, middle, and southern Zhejiang, extending to depths of 53 m, with high-abundance areas concentrated in coastal waters.
Limitations and Future Work
- Limited Environmental Factors: The study acknowledges that while water depth, temperature, salinity, and dissolved oxygen are important, other factors like food availability and ocean currents were not included in the model due to data acquisition limitations. Future work could incorporate these additional variables to improve model accuracy and prediction.
- Overfitting Potential: The authors note that RF models can sometimes exhibit overfitting, although their validation methods aimed to mitigate this.
- Population Differentiation: The study suggests potential genetic differentiation between northern and southern populations of C. lucidus in Zhejiang. Further research could explore this in more detail to understand if management strategies need to be population-specific.
- Data Acquisition Improvement: The authors emphasize that with continuous improvements in data acquisition, incorporating more environmental and biological factors will be crucial for enhancing the fitting and prediction accuracy of future models.
Summary updated