A long-term satellite-based analysis was performed to assess the impact of environmental factors on cyanobacterial harmful blooms (CyanoHABs) dynamics in a typical shallow lake, Lake Taihu. A sub-pixel approach (algae pixel-growing algorithm) was used with 13 years of MOderate-resolution Imaging Spectroradiometer (MODIS) data to evaluate changes in bloom extension, initiation date, duration, and occurrence frequency before and after a massive bloom event (2007). Results indicated that the conditions after this event changed, with a general delay in bloom initiation and a reduction in bloom duration. The environmental drivers of daily, monthly and inter-annual CyanoHABs dynamics were analyzed by detrended correspondence analysis, principal components analysis and redundancy analysis. This demonstrated that wind speed was the main driver for daily CyanoHABs dynamics, and CODmn, total phosphorus and water temperature were closely related to monthly CyanoHABs dynamics. For the year scale, Tmean and nutrients were the main drivers of CyanoHABs initiation date and duration, and meteorological factors influenced CyanoHABs frequency for the whole lake. Regular monitoring of CyanoHABs by remote sensing has become a key element in the continued assessment of bloom conditions in Lake Taihu, and nutrient reduction policies contribute to decrease CyanoHABs occurrence.
Over the past three decades, significant advances in technology and algorithm development have enabled the use of satellites to monitor water quality and algal blooms in coastal seas and inland lakes (Hu 2009; Wang et al. 2011; Palmer et al. 2015), including several attempts to monitor and delineate CyanoHABs and the most common monitoring on intense scum-forming blooms (Gons et al. 2005; Hunter et al. 2008; Wynne et al. 2010). Integrating the spatial and temporal resolutions of the satellite data, significant progress has been made in real-time bloom monitoring of coastal and inland waters especially using the MODerate resolution Imaging Spectroradiometer (MODIS) data (250 and 500 m) (Hu et al. 2010a; Wynne et al. 2013; Kahru & Elmgren 2014; Mouw et al. 2015).
Different variable atmospheric conditions are the first challenge to achieving long-term datasets of algal bloom distributions of inland turbid lakes from MODIS images. To overcome these optical challenges, floating algae index (FAI) was found to be relatively stable and can minimize atmospheric effects (Hu 2009). To date, several studies have been conducted on CyanoHABs dynamics in Lake Taihu by using FAI (Hu et al. 2010c; Duan et al. 2014a; Huang et al. 2014). The second challenge is the coarse spatial resolution of MODIS, where sub-pixel CyanoHABs coverage is necessary to determine accurate spatial coverage. However, MODIS sub-pixel coverage approaches have not been used and can reduce bloom size estimates (for blooms > 100 km2) by 10%–30% (Hu et al. 2010c). By taking into account partial coverage, better early warning of the bloom initiation may be achieved. In addition, the drivers of CyanoHABs in Lake Taihu have been discussed in many studies (Hu et al. 2010a; Zhang et al. 2012a; Duan et al. 2014b) using bivariate comparisons between bloom characteristics (area, initiation date and duration) and economic or environmental factors. Their results suggested that the long-term bloom patterns are driven by both nutrients and climatic factors. However, the drivers of CyanoHABs will depend on the specific temporal scale used, hourly, daily, monthly and yearly. This is the first study to explore the impact of potential drivers on multiple temporal scales.
The present study uses the algae pixel-growing algorithm (APA) (Zhang et al. 2014) to explore the spatial and temporal characteristics of CyanoHABs in relation to catchment and meteorological factors. Environmental driver analysis of daily, monthly and inter-annual CyanoHABs by multivariate methods provided new insights into the environmental drivers of CyanoHABs and lake management at different time scales.
DATA AND METHODS
Bloom coverage determination
The APA was applied to identify the CyanoHABs coverage of every pixel (Zhang et al. 2014) by using the FAI of the central pixel in a 3 × 3 pixel window as a linear composition of the maximum and minimum FAIs of the other eight pixels as follows.
There are three iterative steps in applying the APA to MODIS Rrc data, including calculation of γ for the central pixel, identification of ‘seed’ pixels and decision of the algae coverage using Equation (8). Through iterations, the algal bloom coverage expands from the initial pure algae pixels or high-coverage pixels to low-coverage pixels, according to the relationship between adjacent pixels in a 3 × 3 pixel window. In the end, αMODISpixel is the basic value of algal bloom coverage for subsequent analysis. If αMODISpixel is not zero, the algal bloom area of the pixel is 0.25 × 0.25 × αMODISpixel. For the whole lake, the algal bloom area was calculated by .
Temporal-spatial characteristics of CyanoHABs
In this study, three bloom aspects were identified, namely, the annual initiation date, duration, and the spatial distribution of bloom frequency.
Previous studies indicated that February is the end of the CyanoHABs cycle (Hu et al. 2010c). For each pixel, CyanoHABs initiation was defined as the first moment of non-zero CyanoHABs coverage after February. Similarly, bloom duration was defined as the number of days from the initiation date to the last day when CyanoHABs became zero. For the whole lake, the significant CyanoHABs initiation date and last date were defined as the first and last moments when >25% of the pixels in the lake area showed a non-zero CyanoHABs coverage. The bloom duration of the whole lake was defined from the initiation date to the last date of the whole lake. Initiation dates were transformed into Julian date format.
Meteorological and catchment data
Environmental driver analysis methods of CyanoHABs dynamics
In our research, CyanoHAB drivers at different time scales were analyzed by detrended correspondence analysis (DCA), principal component analysis (PCA) and redundancy analysis (RDA). Firstly, we determined the gradient lengths between characteristics of CyanoHABs and environmental factors at the same time scale, and found that all of the gradient lengths are <2. Then linear models were used to analyze annual, monthly and daily CyanoHABs drivers. Relationships between the daily and monthly drivers were achieved by PCA, and the relationship between annual CyanoHABs characteristics and environmental factors was made by RDA. All of DCA, PCA and RDA calculations were carried out using CANOCO (Ter Braak & Smilauer 2002) computer programs.
RESULTS AND DISCUSSION
Spatial and temporal changes in CyanoHABs area
Spatial and temporal distributions of pixel-bloom frequency
Distributions of pixel-bloom initiation date and duration
Environmental drivers of daily CyanoHABs dynamics in Lake Taihu
Several workers have observed that the formation of near-surface accumulations of buoyant cyanobacteria was dependent on the extent of water column turbulence (George & Edwards 1976; Hunter et al. 2008; Wynne et al. 2010). Cao (Cao et al. 2006) suggested that in Lake Taihu, wind speeds in excess of 4 m/s can sufficiently induce turbulent mixing in shallow lakes and suppress upward migrations through buoyant cyanobacteria, consistently with other studies (George & Edwards 1976). In Lake Taihu, the average wind speed was about 3.5 m/s over the study period, and the wind speed in summer was much lower than that in winter, which favored CyanoHABs formation and duration. Greater than 80% CyanoHABs with >200 km2 took place at low wind speeds (1 m/s to 3 m/s). Given the surface accumulation, vertical heterogeneity can often be compounded by the downwind accumulation of CyanoHABs cells entrained within advective currents, which can lead to patchiness in the spatial distribution of blooms in lakes (Hedger et al. 2002, 2004; Hunter et al. 2008). In Lake Taihu, blooms occurred most frequently with an east–southeast wind, followed by northwest winds. An east–southeast wind favored the accumulation of blooms in the northwest shore and the three bays in the north lake. However, no significant correlation existed between the wind direction and algal bloom coverage.
Environmental drivers of monthly CyanoHABs dynamics in Lake Taihu
Environmental drivers of inter-annual CyanoHABs dynamics in Lake Taihu
Figure 12(a) also indicates that meteorological factors such as Tmin, Tmax, sunshine hours and wind speed affected significant CyanoHABs days and average CyanoHABs frequency of the whole lake. Temperatures from 20 °C to 30 °C favored CyanoHABs, especially the extreme severe blooms, supporting earlier studies by Liu (Liu et al. 2011). Precipitation dynamics are unlikely to directly influence CyanoHABs beyond the reduction in insolation during the rainy season, showing an expected reduction of algal growth and bloom formation, i.e., few extremely severe CyanoHABs occurred between late June and early July.
The biplot of 13 years' CyanoHABs and 10 environmental variables in RDA ordination (Figure 12(b)) demonstrates that CODmn, nutrients and temperature affected CyanoHABs significantly before 2003 and during 2005 to 2007. Wind speed and precipitation were the main drivers in 2003, 2004, 2010 and 2011, while there were no significant environmental drivers of CyanoHABs in the other years.
TWO-SIDE EFFECTS OF HUMAN ACTIVITIES
Although the positive impacts of the actions undertaken are still difficult to detect, a decreasing trend of CyanoHABs occurred despite a growth in population and economic activity. A basin-wide water quality improvement will require a long-term and large-scale commitment to a reduction in nutrient loads. An interesting impact on bloom conditions can be observed by examining the bloom dynamics in Gonghu Bay following the activation of the Water Transfer Engineering from Changjiang River in 2004. The transfer of river water to this eutrophic bay coincided with a reduction in algal bloom frequency with respect to similar areas of the lake (Zhushan Bay, Meiliang bay).
Regular monitoring of CyanoHABs by remote sensing allows a macroscopic and cost-effective assessment of the possible impacts of short-term events (e.g., climate-related) and long-term management and mitigation activities. From the present analysis (2001 to 2013), a decreasing trend was identified after 2007. CyanoHABs develop in nutrient-rich and warm water with low winds and high light availability. The present study indicates that wind speed was the main driver for daily CyanoHABs dynamics. On a monthly basis, CODmn, TP and water temperature were strongly correlated to monthly CyanoHABs dynamics. On an annual scale, Tmean and nutrients appear to be the main drivers of CyanoHABs initiation and duration, and meteorological factors showed key impacts on significant CyanoHABs days and average bloom frequency. On the whole, excessive nutrients and unusually warm temperature caused by climate change were shown to be the main drivers for CyanoHABs dynamics over an extended time scale.
After 2007, favorable meteorological conditions (e.g., continuous low wind speeds) were not accompanied by increased bloom events. We hypothesize that the positive contributions from reduced nutrient loads and biomass reduction in the lake are evident. Ongoing monitoring using MODIS and other satellite sensors will provide additional data to test our hypothesis.
This work was supported by the State Key Program of the National Natural Science of China (Grant No. 41431176) and the National Natural Science Foundation of China (Grant No. 41471287), the National High Technology Research and Development Program of China (2014AA06A509), and the ESA-MOST (China) Dragon 3 Cooperation Program. Monthly water quality data during 2001 to 2013 were provided by the Taihu Basin Authority.