Forest effects on runoff under climate change in the Upper Dongjiang River Basin: insights from annual to intra-annual scales

The Dongjiang River is one of three main tributaries of the Pearl River Basin in southern China (figure 1). The upper reaches of the Dongjiang River Basin (upstream of the hydrometric gauge at Longchuan) have a drainage area of 7932 km². Water availability in the Dongjiang River Basin is key to sustainable social and economic development of the East Wing of Guangdong-Hong Kong-Macao Greater Bay Area, which comprises the cities of Guangzhou, Shenzhen, Dongguan, Huizhou, and Hong Kong Special Administrative Region.

Zoom In Zoom Out Reset image size

The forest coverage in the UDRB has increased rapidly since the afforestation program was implemented in Guangdong Province. During the period 1989–2009, the forest coverage in the UDRB increased from 51.0% to 63.3% (table 1) (Peng et al 2014). On this basis, the study period of 1961–2010 was divided into a reference period of 1961–1990 and an afforestation period of 1991–2010 for comparative analyses in our study.

Monthly averaged discharge data for the period 1961–2010 were obtained from the hydrometric gauge at Longchuan, then monthly averaged runoff time series were generated for the UDRB (upstream of Longchuan). The annual averaged runoff over the UDRB for the study period is 803 mm yr⁻¹. Monthly precipitation datasets, gridded using the methods outlined in Wu and Gao (2013), were based on an interpolation from over 2400 observation stations in China. The annual averaged precipitation over the UDRB for the period is 1624 mm yr⁻¹. Monthly averaged PET estimates were extracted from the Global Land Data Assimilation System forcing data provided by the United States National Aeronautics and Space Administration (Rodell et al 2004). The annual averaged PET over the UDRB for the study period is 1407 mm yr⁻¹.

2.3.1. Process-based method

The process-based method adopted in this study is based on the conceptual rainfall-runoff (CRR) model called SIXPAR, which uses two soil layers to represent the subsurface. The upper zone extends from the surface to the bottom of rootzone, while a lower zone represents ground water storage. The percolation process links the upper and lower zones, simulating the effects of gravity and downward suction (Brazil and Hudlow 1980). In this study, we used a modified SIXPAR (Gupta and Sorooshian 1983, Duan et al 1992) to simulate river runoff at monthly time steps, with precipitation as the input to the upper zone. The actual evapotranspiration is calculated based on maximum soil water capacity, actual soil water content and PET (Zhao 1992, Li et al 2020b). In this study, the model was further modified to include a linear reservoir scheme for routing different runoff components (Zhao 1992). Therefore, the model parameters include: upper and lower zone maximum storage capacities (UM and LM), upper and lower zone recession constants (UK and LK), constants to fit the percolation equation (A and X), and upper and lower zone linear reservoir constants (UR and LR).

In this study, the Multi-objective Shuffled Complex Evolution Metropolis (MOSCEM) algorithm was used for model calibration (Vrugt et al 2003). This algorithm is an improvement over the Shuffled Complex Evolution Metropolis (SCEM) global optimization algorithm, which uses an improved concept of Pareto dominance to evolve the initial population of points toward a set of solutions stemming from a stable distribution (Pareto solution set) (Shi et al 2008). The MOSCEM algorithm is used to calibrate eight model parameters as mentioned above (as shown in table 2). The criterion used for model calibration and validation to ensure an optimal model performance with respect to both observed annual and monthly time series of runoff is based on the Nash–Sutcliffe efficiency (NSE) coefficient (Nash and Sutcliffe 1970), defined as:

$$\begin{equation}{\text{NSE}} = 1 - \frac{{{{\sum}}{{({R_{{\text{obs}}}} - {R_{{\text{sim}}}})}^2}}}{{{{ \sum }}{{({R_{{\text{obs}}}} - \overline {{R_{{\text{obs}}}}} )}^2}}}\end{equation} \tag{ 1 }$$

Table 2. Testing significant differences in precipitation, PET and runoff from annual to monthly scale between the reference period and afforestation period (when Z statistic >1.64 or <−1.64, the significant level is p < 0.05 (one sided)).

Differences between the two periods	Mann–Whitney U-test Z statistic
	Precipitation	Potential evapotranspiration	Runoff
Annual	−0.17	−0.70	−0.60
January	−0.98	0.88	−2.05 a
February	−0.74	−0.24	−2.96 a
March	−0.57	−0.34	−1.07
April	−0.44	−0.68	−0.39
May	2.32 a	−1.38	2.08 a
June	−0.69	1.33	1.16
July	−0.40	−1.04	−1.11
August	−0.96	0.04	−0.60
September	0.13	−0.91	−1.00
October	1.94 a	−2.17 a	−0.18
November	1.29	−1.20	−0.25
December	−0.99	−0.67	−0.94

^a Significant at p = 0.05.

with simulated (R_sim) and observed runoff (R_obs) for the calibration period of 1961–1980, and the validation period of 1981–1990, respectively.

Then, the relative contributions of forest and climate change to annual/monthly averaged runoff were estimated. The total change in annual/monthly averaged runoff can be defined as:

$$\begin{equation}\Delta R = \Delta {R^{\text{c}}} + \Delta {R^{\text{f}}}\end{equation} \tag{ 2 }$$

where ΔR for the total change in annual/monthly averaged runoff can be calculated as:

$$\begin{equation}\Delta R = {R_{{\text{oa}}}} - {R_{{\text{or}}}}\end{equation} \tag{ 3 }$$

where R_oa and R_or represent the observed annual/monthly averaged runoff in the afforestation period and reference period, respectively. ΔR^f and ΔR^c represent the change in annual/monthly averaged runoff attributed to forest and climate change, respectively. As the model parameters are invariant through the reference period and the afforestation period, climate change would take effect solely on runoff in model simulation, and ΔR^c can be estimated as:

$$\begin{equation}\Delta {R^{\text{c}}} = {R_{{\text{sa}}}} - {R_{{\text{sr}}}}\end{equation} \tag{ 4 }$$

where R_sa and R_sr represent the model simulated annual/monthly averaged runoff in the afforestation period and the reference period, respectively. Once ΔR and ΔR^c are estimated, ΔR^f can be calculated by equation (2).

2.3.2. Relation-based method

On the basis of phenomenological considerations, Fu (1981) described the Budyko hypothesis as partial differential equations. Through a combination of dimensional analysis and mathematical reasoning, one analytical solution for annual averaged evapotranspiration was derived (Fu 1981, Zhang et al 2004, Yang et al 2008), and has been analyzed and validated using globally published data (Zhou et al 2015, 2018). Based on the theoretical study by Fu (1981), the dependence of the ratio of annual averaged runoff to annual averaged precipitation (R/P) on the wetness index (P/PET) and watershed characteristics (m) can be estimated as:

$$\begin{equation}\frac{R}{P} = {\left(1 + {\left(\frac{P}{{{\text{PET}}}}\right)^{ - m}}\right)^{\frac{1}{m}}} - {\left(\frac{P}{{{\text{PET}}}}\right)^{ - 1}}\end{equation} \tag{ 5 }$$

which also can be written as:

$$\begin{equation}R = P{\left(1 + {\left(\frac{{{\text{PET}}}}{P}\right)^m}\right)^{\frac{1}{m}}} - {\text{PET}}\end{equation} \tag{ 6 }$$

with R for the annual averaged runoff, P for the annual averaged precipitation, PET for the annual averaged potential evapotranspiration, and m for the watershed characteristics. The values of m can be determined by the observed R, P, and PET in the reference period. Similar to the process-based method, the relative contributions of forest and climate change to annual averaged runoff can be estimated using equations (2)–(4). The relation-based method is used here at annual scale for mutual corroboration with the process-based method.

In figure 2, the left and middle series of panels show the spatial distribution of the seasonal averaged precipitation of the UDRB in the reference period and the afforestation period, respectively. The precipitation in spring (March–May) and summer (June–August) is much greater than that in autumn (September–November) and winter (December–February). The right panel in figure 2 presents the spatial distribution of changes in seasonal averaged precipitation from the reference period to the afforestation period. It clearly shows that precipitation in spring decreased notably, while precipitation in summer showed an obvious increase. In comparison, precipitation in autumn and winter did not change much. In figure 3, the upper and lower panel present the annual time series of precipitation and monthly averaged precipitation in the UDRB. The annual averaged precipitation was 1630 mm yr⁻¹ in the reference period and 1614 mm yr⁻¹ in the afforestation period, respectively. That means the annual averaged precipitation did not change much in the UDRB. When looking into the monthly scale, however, the intra-annual distribution of precipitation showed notable difference in these two periods. Specifically, the seasonal averaged precipitation in spring (March–May) decreased from 625 mm season⁻¹ in the reference period to 582 mm season⁻¹ in the afforestation period, while the seasonal averaged precipitation in summer (June–August) increased from 576 mm season⁻¹ in the reference period to 645 mm season⁻¹ in the afforestation period.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

In figure 4, the left and middle panel show the spatial distribution of seasonal averaged PET of the UDRB in the reference period and afforestation period, respectively. The right panel in figure 4 indicated the spatial distribution of changes in seasonal averaged PET from the reference period to the afforestation period. We found that the PET did not change much for most of the year, except that PET in autumn showed a slight increase. This was also evident from figure 5, the upper and lower panels of which present the annual time series of PET and monthly averaged PET in the upper reaches of DRB.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

According the results from Mann–Whitney U test, there was no significant (p < 0.05) difference in annual averaged precipitation, PET or runoff between the reference period and afforestation period (table 2). Our results also highlighted the differences at monthly scale. To be specific, the precipitation had significant differences in May and October, while the runoff had significant differences in January, February and May. Besides, the PET had significant differences only in October.

Annual and monthly NSE were used for model calibration to ensure an optimal model performance with respect to both observed annual and monthly time series of runoff. Figure 6 illustrates the concept of Pareto dominance to evolve the initial population of points (grey dots) toward a set of solutions stemming from a stable distribution (Pareto solutions, black dots), with respect to observed annual and monthly time series of runoff (annual and monthly NSE). The resulting Pareto sets are then selected as the optimal solutions for further evaluation and analysis. The selected parameters are listed in table 3.

Zoom In Zoom Out Reset image size

The simulated uncertainty ranges of runoff (shaded area) associated with the Pareto solution set estimated using the MOSCEM algorithm are displayed in figure 7, with the observed runoff are indicated with solid line for comparison (Vrugt et al 2003). As mentioned above, the model parameters (representing the underlying surface characteristics) in the process-based method were kept invariant through the reference period and afforestation period, in order to separate the contributions of forest and climate change to runoff change. From the simulation results (table 4), we can see that the model consistently had a better performance simulating the runoff changes in the reference period (with higher NSE values) than in the afforestation period (with lower NSE values). The difference of NSE values between the two periods witnessed that the forest had exerted a great influence on the hydrologic cycle in the UDRB, through altering the underlying characteristics in response to the changing environment.

Zoom In Zoom Out Reset image size

As listed in table 5, the estimated contribution of climate change to annual averaged runoff change from the reference period to the afforestation period ranged from −50 mm yr⁻¹ to −25 mm yr⁻¹, as compared with +21 ∼ +47 mm yr⁻¹ attributed to forest effects in the UDRB based on the results of process-based method. The relation-based method (parameters listed in table 2) provided similar estimates, with the estimated contribution (−21 mm yr⁻¹) of climate change to annual averaged runoff change from the reference period to the afforestation period, compared with +18 mm yr⁻¹ attributed to forest effects. Both methods indicated that forest had a positive effect while climate change exerted a negative impact on runoff in the UDRB. As a result, the observed annual averaged runoff only showed a little reduction (−3 mm yr⁻¹) from the reference period to the afforestation period.

Furthermore, the process-based method could help us look inside the forest effects at the intra-annual scale. The estimated contributions of forest and climate change to monthly averaged runoff change from the reference period to the afforestation period based on the process-based method results are listed in table 6. The estimated contribution of climate change to monthly averaged runoff basically followed the pattern of precipitation change. Except in July and August, climate change exerted negative or little impact on runoff during the year. The spring (March–May) became the season in which the runoff experienced the largest negative impact by climate change. In comparison, the estimated contribution of forest effects to monthly averaged runoff showed a totally different pattern. Except in May and June, forest exerted positive or little impact on runoff during the year, while the winter (December–February) experienced the largest positive impact on runoff. As a result, the observed monthly averaged runoff in May and June experienced notable reduction, while those in other months experienced increase or no change. The UDRB provides evidence that additional forest cover would not injure but even increase runoff, especially dry season runoff.

This study provided an insight of climate change and large-scale afforestation impacts on runoff from annual to intra-annual scale in the UDRB during the past decades. Our results indicated there was almost no change in the annual averaged precipitation. However, precipitation in spring decreased notably, while that in summer showed an obvious increase from the reference period to the afforestation period. This finding agreed with previous studies (Zhou et al 2010, 2016). The reasons for the precipitation change might be complicated. It is well known that the climate has been changing with the changed hydrologic cycle, including the precipitation patterns globally and regionally (Ellison et al 2017). In addition, forest also plays an active role in regulating atmospheric moisture fluxes and precipitation patterns (Ellison et al 2017). Our results also indicated that the PET did not change much during the year, except for a slight increase in autumn, similar to the findings in Zhou et al (2010). We also understand that forests play a large role in regulating fluxes of atmospheric moisture and rainfall patterns over land (Ellison et al 2017). However, as this issue is well beyond our research scope, as well as our framework and methodology, we would rather attribute the precipitation change to the changing climate and put the focus on the direct forest effects on runoff.

Climatic variability and land cover/use changes are commonly recognized as two major drivers for hydrologic regimes. The methods, namely process- and relation-based methods, were used in this study to separate the impacts of forest and climate change on runoff. The relation-based method derived from Fu's equation (Fu 1981), had been confirmed as a valid framework for quantifying the effects of land cover and climate on hydrology (Zhang et al 2004, Yang et al 2008). According to the recent works based on Fu's equation on global runoff pattern (Zhou et al 2015, 2020), afforestation can increase runoff in large watersheds (m > 2) or humid regions (P/PET > 1). The UDRB with a drainage area of 7932 km² (m ≈ 1.86) and P/PET ≈ 1.15 provides evidence that additional forest cover would not injure but even increase runoff at annual scale. At monthly scale, our process-based method results witnessed that afforestation would increase dry season flows where improvements in soil infiltration capacity and groundwater recharge exceed increased evapotranspiration (Bruijnzeel 2004, Zhou et al 2010, Ellison et al 2017).

Regarding to the basins experiencing afforestation/reforestation, the forest effects can be divided into two parts, namely the initial and additional forest coverages. The results of previous studies on the relationship between forest and water are highly variable, especially in the basins experiencing afforestation/reforestation. We found some new findings regarding the initial forest coverage of those basins (table 7). For the basins with smaller initial forest coverage (e.g. lower than 30%), the runoff showed decreasing trends after afforestation/reforestation (Zhang et al 2011, Liu et al 2014, Wang et al 2016). In contrast, the runoff in the basins (e.g. UDRB) with larger initial forest coverage (e.g. higher than 30%) was found with no or increasing trends after afforestation/reforestation (Antonio et al 2008, Zhou et al 2010, Wu et al 2015). We understand that the physiological responses of forest to rising CO₂ may reduce plant transpiration through stomatal closure (Gedney et al 2006, Betts et al 2007, Fowler et al 2019). From a hydrologic perspective at the basin scale, the importance of physiological responses of forest to rising CO₂ on evapotranspiration and runoff directly depended on the initial forest coverage, and potentially explained the different runoff responses of those basins experiencing afforestation/reforestation. To be specific, as climate change progresses and CO₂ rises, the basins with larger initial forest coverage would result in more transpiration reduction from the initial forest cover than the basins with smaller initial forest coverage, then would be more likely to offset or even surpass the transpiration increase from the additional forest cover and sustain more runoff in comparison.

The effects of forest on runoff have been debated for a long time. This issue would become more complicated involving deforestation, afforestation or reforestation. Numerous studies have been conducted throughout the world to address this concern. By using a combination of process- and relation-based methods, this study revealed that the balance of forest and climate change has resulted in a slight reduction in runoff in the UDRB over the past decades. Our results provided new evidence that additional forest cover would not injure but even increase runoff, especially dry season runoff (Sun et al 2006, Ellison et al 2017, Bai et al 2020). Limited to the framework and methodology, our study attributed the precipitation change completely to the changing climate and cut off the linkage between the forests and the precipitation. However, forests-driven evapotranspiration contributes to the availability of atmospheric moisture vapor and its transport locally and regionally, raising the likelihood of precipitation events and increasing runoff (Ellison et al 2012, 2017). And this mechanism is becoming more complicated under the changing climate and rising CO₂ concentration (Fowler et al 2019). Therefore, more efforts are needed to improve the understanding of water-energy balance and associated biophysical processes between land–atmosphere interactions in forest-dominated watersheds.