4.1.1. Accuracy Assessment

Previous studies have proved that the support vector machine (SVM) method has high interpretation accuracy in fish pond extraction [22]. This could also be seen in Figure 13A that the water body delineation using SVM was satisfactory.

Sometimes, problems happened to some narrow ponds with width less than 30 m, which could be false to be identified in Landsat imagery due to relative coarse spatial resolution (Figure 14) [23]. However, in consideration of the satellite return interval, Landsat satellites were the best choice. Table 2 has shown that the Kappa coefficient for each yearly classification results were all above 0.8, capable for the data analysis. To maintain the data quality, visual interpretation was combined to improve classification results; ye<sup>t</sup> the result may still be affected by mixed pixels due to images' resolution (Figure 13B).

In addition, it should be noted that, because the study area is located in the subtropical area, there are almost no cloud-free satellite images available in the monsoon season. Most of the images used in this study were acquired during dry seasons. Generally, the surface area of the water bodies varies greatly during from monsoon season to the dry season, and the area during the dry season is commonly significantly smaller than in the monsoon season. However, as reported in Section 3, most of the fish ponds are now artificial, and changes in water surface area are rarely affected by season switch. Therefore, this impact can be safely ignored in this study. Besides, for a time scale of over 40 consecutive years, the Kappa coefficient of over 0.8 is well enough to perform a relatively accurate change trajectory of fish ponds in such a large scale of this study. However, it should be noted that, when similar methods are used to evaluate natural water bodies, this effect should be considered.

**Figure 13.** Image classification results and the subsequent visual interpretation results: (**A**) the classification results using SVM; (**B**) the results of visual interpretation.

**Figure 14.** Fish ponds in Google Earth high resolution image (**A**) and Landsat image (**B**).



### 4.1.2. Comparisons with Previous Studies

Previously, few studies were conducted on fish ponds in the entire Guangdong–Hong Kong–Macao Greater Bay Area. Many studies have been conducted in a specific city or a smaller area. Therefore, in the comparison, previous study results were compared with our results from the same specific areas. Previous studies about trends of dyke-ponds between 1978 and 2016 in Shunde, a district of Foshan City located in central GBA, reported that the spatial distribution of fish ponds in Shunde did not greatly change during 1988 and 1993 [24]. It was concluded from the comparison of fish pond distribution between these two years. Completing the gap of the time series by adding the fish ponds distribution in 1991 (Figure 15), we found that during this period, fish ponds in the eastern Shunde experienced a dramatic decrease, lost a large number of ponds, most of which disappeared in 1991.

**Figure 15.** Spatial changes in fish ponds in Shunde District between 1988 and 1994.

Another previous study by Li [11] displayed the distribution changes of fish ponds in 1964, 1976, 1988, 2000, 2012 in Pearl River Delta region. The results indicated that fish ponds kept increasing in its area from 1964 to 2012 [11]. This study also covered the period of 1988 to 2012. However, due to some cities' boundaries adopted by the two studies are slightly different, resulting in false comparisons for such cities. The boundary of Guangzhou that two studies adopted are relatively consistent. Li's study displayed that the classification results of 1988, 2000, 2012, were 41.17 km2, 203.06 km<sup>2</sup> and 213.44 km2, respectively. Comparing with this study's results in 1988, and 1999 and 2013 (no results available for the years of 1999 and 2012), which are 45.86 km2, 190.56 km<sup>2</sup> and 228.62 km2, respectively, the two results are basically consistent, with an averaged relative difference of 10.93%. However, it should also be known that this difference may also be caused by the real difference in the latter two periods (1999 vs. 2000, 2012 vs. 2013). Despite the difference, the time series in our study more thoroughly demonstrate the trend of fish pond

changes in the study area between 1988 and 2013 (Table 3). In addition, the obtained trends are quite consistent, but Li's study used merely 3-year data over a 24-year period, resulting in a harder response in sensitivity. This study revealed that fish ponds in Guangzhou actually experienced a fluctuation instead of a smooth increase, especially in the period of 1994 to 2013. Besides, the expansion of fish ponds happened before 2012.

**Table 3.** Comparison of the pond changes between Li's study and this study in Guangzhou for the period 1988 to 2013.


### *4.2. Possible Causes for Pond Changes in Different Cities and Future Speculation*

### 4.2.1. The Growing Cities

Guangzhou, Huizhou, Zhuhai, Zhongshan, Zhaoqing, Jiangmen—six cities in total were classified as growing cities. The reason was that compared with 1986, all of these cities experienced an increase in pond area in 2019. The peaks also appeared before 2019, showing a fluctuating growing trend.

Guangzhou's population increased by 1.67 million between 1990 and 2006, about 28% in total, ye<sup>t</sup> the actual product consumption was about 3.7 times of the previous amount. The aquatic products in 2006 was 4.9 times of that in 1990 [22]. The inflow of population and the expanded market stimulated the demands and development of aquaculture, resultant increase in fish ponds in Guangzhou before the beginning of the new century, especially in the Nansha District of Guangzhou. Therefore, Nansha District experienced a strong expansion from 1986 to 2013. After 2013, Nansha District was designated as a free trade zone, and Nansha Port was also developed as a manufacturing and industrial export, resulting in further changes in land uses and shrinking in fish ponds [25]. Huizhou was similar too. The area of fish ponds near the seaside also maintained a positive growth before the development of the petrochemical industry before 2006, and then gradually shrank with industrial development [26].

Zhaoqing City also achieved rapid growth in fish ponds from 1986 to 2006 through the reconstruction of low-lying sandy wasteland [27]. Subsequently, from 2006 to 2009 and 2013 to 2019, due to the impact of natural disasters, bacterial diseases, and dramatic price fluctuation in aquatic products, the risk of aquacultural development increased, which dampened the enthusiasm of farmers for aquacultural development, and the area of fish ponds showed a rapid downward trend.

Driven by the adjustment of the regional agricultural policy, the area of fish ponds in Zhuhai had increased significantly before 2006. During the period 1990–2006, local cultivated land experienced an accelerated loss, during which the net transfer area of cultivated land to the fish ponds reached 16,054.23 hm<sup>2</sup> [28,29]. Subsequently, urbanization developed during 2006–2009. As there were less cultivated land and forest land available for urbanization, the occupation of fish ponds was accelerated to a certain extent. The Government planned to accelerate the development of ecological fisheries and the construction of agricultural and fishery infrastructure to provide a guarantee for the development of fish ponds. Similarly, the Local Government of Jiangmen released a new policy [30] in 2009 to prompt aquacultural development the growth of the fish ponds in Jiangmen from 2009 to 2013.

However, from 2013 to 2019, Zhuhai has invested heavily in the development of ecological agriculture such as flowers, fruits and vegetables, and organic rice. The Government

introduced modern agricultural industrial parks, such as the Yongcheng Horticulture and Taiwan Orchid Greenhouse Planting Base, and eliminated fish ponds with low economic and environmental benefits, causing the shrinking of fish ponds.

### 4.2.2. The Shrinking Cities

The fish ponds in Foshan and Dongguan showed an initial increasing trend followed by a significant shrinking. From 1986 to 1994, the area of fish ponds in Foshan City has increased significantly. Taking the Nanhai County in Foshan City as an example, the county became a pilot for land policy reform in 1987. Stimulated by the flexible land policy, the productivity has greatly increased. The traditional agriculture was drastically reduced and transformed into vegetable planting and aquaculture. After 1994, Nanhai county was cancelled and became a district of Foshan. In order to promote industrialization, the Government reduced the rent of collective land in rural areas to attract a large number of enterprises to settle in. The upgradation in the industry prompted a rapid expansion of built-up land, with an average annual growth rate of over 7% [31]. In addition, township and village enterprises had sprung up all over Foshan. In 1991, the total income of township and village enterprises in Foshan reached 21.04 billion Chinese yuan, which increased by 3.84 times in 1997 [32]. The expansion of built-up land encroached on a large amount of fish ponds, resulting in shrinkage of fish ponds and its fragmented distribution. However, in the most recent 5 years, a series of development policies, such as the "Conservation and Development Plan for the Agricultural System of Fish Ponds in the Pearl River Delta, Foshan, Guangdong" and "Strategic Plan for the Implementation of Rural Revitalization in Guangdong Province (2018–2022)" [33] have slowed down the shrinking. At the same time, the Foshan Local Government has declared the fish ponds as an important agricultural cultural heritage in China, and combined it with the tertiary industry, which had a certain effect on protecting the fish ponds and increasing their outputs. Such measures have slowed down the shrinking trend from 2013 to 2019, but remained insufficient in reversing the shrinking trend.

### 4.2.3. The Fast Shrinking Cities

The number of fish ponds in Hong Kong SAR and Shenzhen was relatively small, but they have shrunk at a rapid rate (Figure 16). The fish ponds in Hong Kong are mainly located on the river alluvial plains in the estuary of the Shan Pui River in Yuen Long and Nan Sang Wai. In the 1980s, the development of fish ponds in Hong Kong reached its peak. However, with the rapid development of Yuen Long after the 1990s, the land used for the fish ponds in Nan Sang Wai had changed. For example, the original fish ponds were excavated and converted to a new channel of the Kam Tin River or recreated as smaller triangular ponds near the river channel. Moreover, the discharge of wastes in the process of urbanization had led to the continuous deterioration of the water quality of Shan Pui River, which is not suitable for aquaculture now, resulting in the disposal of local fish ponds.

The fish ponds in Shenzhen are mainly concentrated in the original Bao'an County. After the establishment of the Shenzhen Special Economic Zone in 1980, superior policies and other geographical conditions promoted the rapid development of industry. The demand for built-up surged, and changes in land uses led to a sharp decline in fish ponds over the past 40 years. Therefore, the urbanization is the major cause for fish pond shrinking in Shenzhen and Hong Kong. Besides, large amounts of aquatic products imported from mainland China was also an important cause for the shrinking of fish ponds in Hong Kong.

**Figure 16.** The rapid shrinkage of fish ponds in Shenzhen and Hong Kong SAR in the 1990s.

### 4.2.4. Future Trends of Fish Ponds in GBA

In view of the current trends of changes in various cities and the overall planning background of the Greater Bay Area, the shrinkage of fish ponds in the future will remain for a long time. However, the rate of shrinkage in various regions will vary greatly due to various drivers such as local development policies. It is expected that the development of fish ponds will tend to integrate with the tourism and service industry. The economic benefits of fish ponds will increase accordingly via excavation of the cultural value of fish ponds and construction of traditional aquacultural demonstration areas. Due to the differences in regional development and local governmen<sup>t</sup> investment, the better evolution in fish ponds could appear in such regions where the economic development is quite high.
