Climate-driven land surface phenology advance is overestimated due to ignoring land cover changes

Global warming has led to earlier spring green-up dates (GUDs) in recent decades with significant consequences for global carbon and hydrologic cycles. In addition to changes in climate, land cover change (LCC), including interchanges between vegetation and non-vegetation, and among plants with different functional traits, may also affect GUD. Here, we analyzed how satellite-derived GUD from 1992 to 2020 was impacted by changes in temperature, precipitation, standardized precipitation evapotranspiration index (SPEI), solar radiation, and LCC for the Northern Hemisphere (>30° N). While the climate variables had larger impact overall, variability in GUD was controlled by LCC for 6% of the Northern Hemisphere, with systematically earlier or later changes among transitions between different land cover types. These changes were found mainly along the southeastern coast of the United States, in Central-north Europe, and across northeastern China. We further showed that climate change attribution of earlier GUD during 1992–2020 was overestimated by three days when the impact of LCC was ignored. Our results deepen the understanding of how LCC impacts GUD variability and enables scientists to more accurately evaluate the impact of climate change on land surface phenology.


Introduction
Plant phenology, especially spring green-up date (GUD), is highly sensitive to climate change (Richardson et al 2013) yet also impacts climate and terrestrial ecosystems processes, forming a feedback loop (Richardson et al 2013, Piao et al 2019a, 2019b. Consequently, research has increasingly focused on the effects of climate change on phenology (Richardson et al 2013, Piao et al 2019a. Advancing GUD based on satellite data has been revealed by recent researches, which is closely related to warming, light period, and CO 2 fertilization (Richardson et al 2013, Piao et al 2019a. Numerous studies of both in situ and satellite-derived phenology have shown that the earlier onset of GUD in the Northern Hemisphere is, in part, a result of global warming (Forkel et al 2014, Keenan 2015. The complex relationship between GUD and climate variables has been observed. For example, preseason maximum temperature and winter precipitation had great influence on GUD (Piao et al 2015), chilling days and light period also had a considerable impacts on GUD trend (Fu et al 2015). In addition, except climate change, the GUD trend also varied at different altitudes (Vitasse et al 2018). The causes of GUD changes are complex that include climate change and land cover change (LCC). LCC can change the energy balance and biogeochemical cycles, further affecting surface characteristic (such as GUD) (Duveiller et al 2018). Therefore, quantifying changes in phenology and LCC impacts is of prime importance when analyzing climate change impacts and accurately estimating ecosystem carbon flux.
Mechanisms surrounding this trend in GUD are complex. The GUD is closely related to temperature, precipitation, and radiation (Richardson et al 2013, Piao et al 2019a. Warmer spring temperatures decrease the amount of time required to meet a species' growing degree day (GDD) requirement for green-up to begin. However, temperature effects are strongly modulated by winter precipitation (Forkel et al 2014, Fu et al 2014, Yun et al 2018. An increase in winter precipitation falling as snow may increase the time it takes vegetation to meet the GDD requirement, especially in temperature-limited ecosystems (Yun et al 2018). Solar radiation also influences GUD because it partly represents photoperiod (daylength) (Richardson et al 2013, Piao et al 2019a and itself may be affected by precipitation (Tang et al 2016). For example, more solar radiation usually means higher surface temperature and a long photoperiod, and promotes earlier GUD (Richardson et al 2013, Tang et al 2016. In addition, climate change impacts on GUD vary with elevation, vegetation type, and tree age. Changes in advancing GUD vary with elevation as stronger trends are found at higher elevation, likely caused by faster preseason warming in these locations over time (Piao et al 2011, Vitasse et al 2018. Despite considerable effort, the mechanistic understanding of GUD dynamics and its drivers is incomplete. In addition to the parameters listed previously, LCC exerts a significant impact on land surface phenology change. Grasslands usually have earlier GUD than forests because grasslands require fewer GDD than forests (Ganguly et al 2010, Jeganathan et al 2014, while the GUD for trees becomes later as they age because older trees require more GDD than younger ones (Menzel and Fabian 1999). Case studies on intensive agricultural areas (Zhang et al 2019) and burned forest areas Zhang 2017, 2020) have highlighted the essential role of LCC in understanding the spatial and interannual variations in land surface phenology widely associated with climate change. Studies have shown that the long-term trend in GUD for intensively cultivated areas is influenced by LCC and climate change, and the influence of LCC on GUD dominates in some regions (Zhang et al 2019). For example, the GUD was delayed before an area burned but that now it occurs earlier and has advanced by ∼15 days (Wang and Zhang 2017). However, the quantitative contribution of LCC to changes in GUD has not been systematically studied.
The earth's land surface has changed as a result of urbanization, afforestation, and cropland abandonment in the Northern Hemisphere during the past few decades and similar changes will continue into the future (Winkler et al 2021). Satellite-derived phenology provides an indication of current climate change from regional to global scales (Piao et al 2019a. In particular, satellite-derived GUD have been widely employed to investigate warming impacts on promoting earlier GUD because of its wide coverage and long, continuous time-series. However, the spatial resolution of satellite-derived phenology used in global climate change studies is coarse (e.g. 500 m × 500 m or 0.05 • × 0.05 • ). Phenology derived using coarse resolution LCC data may be incorrect because of the mixed pixel effect , Peng et al 2017b, 2018, Chen et al 2018. When analyzing interannual variation in satellite-derived phenology, many factors contribute to the final value assigned to each pixel, including LCC, climate, and other factors. To identify the effects of climate change on changes in GUD, we must first determine the impact of LCC (Richardson et al 2013, Helman 2018, Zhang et al 2019. To determine the impact of LCC, we used long-term satellite-derived GUD and land cover data to characterize the LCC impacts on GUD for the period 1992-2020 in the Northern Hemisphere (>30 • N). We examined the statistical relationships between the GUD and several parameters (temperature, precipitation, radiation, standardized precipitation evapotranspiration index (SPEI) and LCC). Finally, trends in GUD were attributed proportionally to these different drivers.

Satellite-derived GUD
The GUD data were derived from Global Long-Term Climate Modeling Grid Land Surface Phenology (CMGLSP, 1992(CMGLSP, -2016.05 • ) (Zhang 2015), and VIIRS/NPP Land Surface Phenology Collection 2 (VNP22C2, 2013-2020, 0.05 • ) . Phenology metrics from CMGLSP and VNP22C2 were retrieved using a hybrid piecewise logistic model (Zhang et al 2003, 2018, Peng et al 2017c. We validated the consistency between CMGLSP and VNP22C2 for the period 2013-2016 when both datasets were available. We calculated the following to test numeric correlation: root mean square errors (RMSEs), Pearson correlation coefficient (R, ranging from −1 to 1) and agreement coefficient (AC, ranging from 0 to 1) (Ji and Gallo 2006). To test for spatial pattern consistency, we used the spatial efficiency metric (SPAEF, ranging from 0 to 1) (Koch et al 2018). The definitions and equations of these indices can be found in supplementary section 1. A higher value of R, AC, and SPAEF indicates a stronger numeric correlation and spatial consistency between the two datasets. The R and SPAEF were both greater than 0.9, AC was greater than 0.85, and RMSE was smaller than 13 days, indicating that numeric correlation and spatial patterns were consistent for the two datasets (supplementary figure S1). As a result, a GUD dataset for the period 1992-2020 was generated by combining CMGLSP for the period 1992-2016 with VNP22C2 for the period 2017-2020 with no further changes.

Climate data
Changes in GUD are highly associated with climate variables from preceding months (Piao et al 2015). We defined preseason as the period from 1 November of the preceding year to mean GUD (1992-2020) (Piao et al 2015). The ERA5-land reanalysis dataset provides a continuous hourly record of global land surface variables at 0.1 • resolution since 1950 (Muñoz-Sabater et al 2021). We converted hourly to daily by taking the maximum of temperature, the sum of precipitation, and the average of radiation, within a day. We selected three climate variables generated by ERA5-land and calculated a preseason value for each year for the following variables: average maximum temperature (TMP, an average of all days maximum temperature, unit: • C), precipitation (PRE, a sum of all days precipitation, unit: mm), radiation (RAD, an average of all days downward radiation, unit: W m −2 ). We also utilized monthly composite ERA5-land precipitation and potential evapotranspiration data to calculate SPEI in order to analyze the impact of drought on GUD. The timescale for SPEI was set at six months, and we chose the SPEI value for the 4th month of each year to represent the preseason drought conditions for that year. Climate variables were resampled to a resolution of 0.05 • using bilinear interpolation to match the resolution of the GUD data.

Land cover data
We used European Space Agency Climate Change Initiative Land Cover (ESA CCI LC) data with 300 m resolution for the period 1992-2020 to investigate the impacts of LCC on the GUD trend. The ESA CCI LC provides continuous and annually updated global land cover products from 1992 to 2020 with 37 United Nations Land Cover Classification System (UNLCCS) classes. These data can be used as inputs for climate models as well as for scientific research such as forest and desertification monitoring and LCC monitoring (Hollmann et al 2013). Errors in land cover classification may be larger than LCC itself. Data production processes for ESA CCI LC ensured that each change persisted for more than two successive years to reduce false change detections (Li et al 2018). Because of irrigation and human management, changes in cropland are difficult to correlate with GUD, so we masked permanent cropland pixels for the period 1992-2020 to reduce error. We calculated a coefficient of variability based on the 37 land cover class fractions within each 0.05 • × 0.05 • grid using equation (1) to represent LCC year by year where S andx are the standard deviation and mean value of the 37 land cover class fractions within a 0.05 • × 0.05 • grid.

Analyses
We used multiple linear regression to explore the effects of TMP, PRE, SPEI, RAD, and LCC on GUD (Le Provost et al 2020). The beta coefficients were interpreted as the single standard deviation change in the dependent variable caused by a single standard deviation change in the independent variable (Le Provost et al 2020). These coefficients were estimated using the ordinary least square method shown in equation (2): where Y is GUD, x i is normalized independent variable, β i is beta coefficient. The relative contribution of each independent variable is defined as the corresponding percent of beta coefficient to the sum of all beta Table 1. Rules to determine the effect of relative contribution. A positive relative contribution indicates that changes of the independent variable causes delaying GUD, a negative relative contribution indicates that changes of the independent variable causes advancing GUD.

Sign of beta coefficient
Sign of Thiel-Sen's slope Sign of relative contribution coefficients. Compare to slope detected by simple linear regression, slope from Thiel-Sen estimator is more robust for effect of outlier removal. Therefore, the sign of relative contribution was jointly determined by the sign of corresponding Thiel-Sen's slope and beta coefficient (table 1). A positive coefficient indicates that an increase in the independent variable delays GUD, while a negative coefficient indicates that an increase in the independent variable advances GUD. We defined the dominant driving factor for each grid as the variable with the largest beta coefficient.
To identify the signal of GUD changes due to each individual LCC from 1992 to 2020, a ridge-regression method was introduced to unravel the effect of each LCC on GUD (Huang et al 2020). To facilitate interpretation of LCC, the 37 UNLCCS classes were aggregated into the International Geosphere-Biosphere Programme (IGBP) classes using the crosswalk established in other studies (Duveiller et al 2018, Huang et al 2020. A set of 5-by-5 moving windows was used to decompose GUD changes resulting from the mix of the possible LCCs. For window i, a model was built using IGBP classes fractions of each of the 25 grids of the window as shown in equation (3) where X i is an explanatory variable matrix containing the fractions of all land cover classes in each of the 25 grids in the window; y i is a vector containing 25 GUD values; β i is the vector of the regression coefficients; and ε i is the vector of the model residual. The model is then solved using ridge-regression. Finally, multiple year mean matrices were obtained for the period 1992-2020 describing changes in land cover transitions and the corresponding standard errors using established methods (Huang et al 2020). The number of GUD changes misinterpreted by LCC was defined as the product of LCC's relative contribution and the change in GUD for the period 1992-2020.

Relative contribution of LCC, temperature, precipitation, SPEI, and radiation to changes in GUD
Between 1992 and 2020 the GUD started 5-15 days earlier for 35% of the area, and it was mainly found in areas above 60 • N and across most of Europe. About 7% of the area in Eastern Europe and Central Russia showed GUD starting 15-30 days earlier. The GUD was delayed for 39% of the land area found in central North America, northern China, and regions between 80 • E-100 • E and 50 • N-60 • N (figure 1).
We estimated LCC as coefficient of variability of land cover class fractions within each 0.05 • × 0.05 • grid, then, linear statistical relationships between GUD and LCC, temperature, precipitation, SPEI and radiation were analyzed (section 2), revealing that the effects of these drivers on GUD varies regionally. The LCC delayed GUD across 50% area, which was mainly found in west-central Russia (50 • E-90 • E, 55 • N-65 • N), the southeastern coast of the United States and the Great Lakes region (figure 2(a)). The relative contribution of LCC ranged from 0% to 20% across 37% of the area ( figure 2(a)). Temperature advanced GUD in 75% area, with 22% of that area having a relative contribution greater than 40% ( figure 2(b)). Only 25% of the area had a delayed GUD response to temperature-primarily in Central America (120 • W-90 • W, 45 • N-60 • N) ( figure 2(b)). Precipitation delayed GUD across more than 60% of the area, mainly above 60 • N, and 25% of the area had a relative contribution between 20% and 40% (figure 2(c)). The area of advanced GUD caused by precipitation was mainly distributed across parts of Central Asia with a dry or semi-dry climate (figure 2(c)). The area of advanced GUD caused by SPEI changes was mainly found in Central Asia and Russia (figure 2(d)). Radiation's spatial contribution was generally opposite that of precipitation. For example, radiation advanced GUD in Northeast China while precipitation delayed GUD (figures 2(c) and (e)).
Based on the annual trends of GUD and various driving factors during different time periods, significant differences were found between 1992-2000, 2001-2010, and 2011-2020. TMP, PRE and SPEI increased most rapidly during 2011-2020, while GUD, LCC decreased most rapidly during 2011-2020. It can be concluded that the period of 2011-2020 was the period during which LCC and other climatic variables had the greatest impact on GUD (see supplementary section 2 and figures S2-S9). After analyzing the impact of LCC, TMP, PRE, SPEI, and RAD on GUD changes along latitude, altitude, and precipitation gradient, several patterns were identified. Specifically, LCC was found to cause GUD  to delay in the latitude range of 30 • -40 • N and to advance in the latitude range of 60 • -70 • N, with a higher relative contribution at altitudes above 4000 m. An increase in SPEI mainly caused GUD to advance when the precipitation was less than 1000 mm, but caused GUD to delay when the precipitation exceeded 2000 mm (see supplementary section 3 and figure S10).

LCC impact on changes in GUD is non-negligible
The factor with the greatest relative contribution to each grid was defined as the dominant factor (section 2), and the area influenced predominately by LCC accounted for 6% of the total study area. The contribution of LCC to interannual variation in GUD was considerable in Northeast China, along the southeastern coast of the United States, and throughout Central Russia. Temperature, precipitation, SPEI, and radiation had greater impacts on interannual changes in GUD overall, being the dominant factor in 38%, 23%, 10% and 23% of the area, respectively (figure 3). Advancing GUD dominated by temperature accounted for about 25% of the area and was mainly distributed above 60 • N, while delayed GUD was mainly found in Central America (

Characteristics of changes in GUD for specific LCC
Using methods from a previous study (Huang et al 2020), we identified GUD changes caused by transition between individual land cover types. We filtered the data for illogical land cover transitions (Peng et al 2017a), such as transformations from deciduous broadleaf forests into evergreen needleleaf forests in a single year. As shown in figure 4, the GUD changes induced by individual LCC varied greatly. Some transition types did not occur (e.g. transition from cropland or urban areas to shrublands), so changes in GUD could not be identified. The transitions from deciduous broadleaf forests, savannas, croplands or wetlands to other land cover types usually caused advancing GUD, with transitions from deciduous broadleaf forests, savannas and wetlands to urban areas advancing GUD by about ten days, eight days, and seven days, respectively. The transition from deciduous needleleaf forests or urban areas to other land cover types usually delayed GUD, with transitions from deciduous needleleaf forests to grasslands and urban areas to sparse vegetation inducing a change of about four days and five days, respectively. Changes in GUD between individual land cover types had opposite responses (Duveiller et al 2018). Therefore, reforestation of deciduous needleleaf forests or urban expansion usually advanced GUD, while deforestation and urban reduction delayed GUD. Similarly, cropland expansion often delayed GUD, while cropland reduction had the opposite effect.
The land cover transition matrix for the period 1992-2020 reveals that about 8% of the area underwent land cover transitions, which is comparable to the 6% of the area whose dominant driver of GUD change was LCC. The two land cover types that decreased most were evergreen needleleaf forests (1.5%) and bare/sparse vegetation (0.99%). The two  land cover types that increased most were savannas (1.53%) and croplands (1.28%). The three land cover transitions seen most often were evergreen needleleaf forests to savannas, grasslands to croplands, and deciduous needleleaf forests to savannas, which were 0.47%, 0.38% and 0.36%, respectively. These three primary land cover transitions delayed GUD two days, three days, and four days per year, respectively, which is one reason why the percent of LCC that delayed GUD was higher than the percent of LCC that advanced GUD.

Climate change impact on advancing GUD was overestimated
The GUD changes induced by LCC are shown in figure 5(a). The average advanced (3.0 days) and delayed (3.3 days) LCC-driven GUD changes from 1992 to 2020 accounted for about 22% and 15% of corresponding GUD changes, respectively (figure 6). The average advanced (13.6 days) climate (including temperature, precipitation, SPEI and radiation)driven GUD changes from 1992 to 2020 accounted for 81% of corresponding GUD changes, respectively Figure 6. The area-weighted average of changes in GUD and the contributions of land cover change (LCC), temperature (TMP), precipitation (PRE), standardized precipitation evapotranspiration index (SPEI) and radiation (RAD) for the period 1992-2020. The unit is days, the blue bar is average of all grids, the red bar is average of all positive grids, and the green bar is average of all negative grids.
(figure 6). The distribution of LCC-driven GUD changes was uniform. Overall, LCC delayed GUD in about 57% of the area, with 6% of the area delayed by more than one to three days. On the other hand, LCC advanced GUD in about 43% of the area, with 12% advancing one to three days (figure 5(a)). Regions with noticeable LCC-driven GUD changes were located mainly along the southeastern coast of the United States, in Central-northern Europe, and in Northeastern China (120 • E-135 • E, 45 • N-50 • N) with about seven days, six days, and nine days advancing GUD, respectively. We selected four sub-regions to verify if GUD changes caused by LCC agreed with the results shown in figure 4. In these four sub-regions, urban and cropland expansion were the major drivers of GUD change. In the sub-regions shown in figures 5(b) and (d), the main LCC was the transition between grassland and cropland, thereby delaying and advancing GUD, respectively. In the sub-region shown in figure 5(c), the main transition was urban expansion (figure 5(g)), which advanced GUD. In the sub-region shown in figure 5(e), the main change in LCC was deciduous needle leaf forest deforestation (figure 5(i)), which delayed GUD.

Validation of calculated GUD trends
In this study, on average, GUD advanced 1.1 days per decade in the Northern Hemisphere from 1992 to 2020. A previous study showed that GUD for the Northern Hemisphere advanced on average 2.1 days per decade for the period 1982-2011 (Piao et al 2019a). A second study revealed that the rate of change in GUD per 1 • C of warming decreased by half when comparing the period 1999-2013 to that of 1980-1994(Fu et al 2015. In addition, advancing GUD trends observed in most of Europe and half of North America (figure 1) were consistent with results of yet another study that used different methods to calculate GUD (Huang et al 2017). In a qualitative assessment, this demonstrates that the GUD data used in this study are able to capture similar phenology dynamics to those identified by others.

The role of climate
Temperature is considered to be a primary factor controlling changes in GUD, and previous studies observed that maximum temperature had a greater influence on GUD changes than minimum temperature (Piao et al 2015, Fu et al 2016, which is why we selected average preseason maximum temperature. We also confirmed that recent warming has led to advancing GUD (Tang et al 2016, Piao et al 2019a ( figure 2(b)), which is shown by the negative correlation between temperature and GUD in the Northern Hemisphere ( figure 7(b)). Precipitation plays a co-dominant role with temperature in land surface phenology dynamics (Forkel et al 2014). At latitudes above 50 • N, increased preseason precipitation usually delays GUD, showing a positive correlation (figures 2(c) and 7(c)). In mountain areas and in cold regions, e.g. continental northern regions (>50 • N), increased precipitation may occur partly as snow (Shutova et al 2006). Increased snow cover in spring may melt later and delay GUD (Shutova et al 2006, Tang et al 2016. On the other hand, increased water supply in spring can advance GUD in warm regions (Shutova et al 2006). Moreover, increased preseason precipitation can be accompanied by more cloud cover, which reduces the amount of incoming shortwave radiation (Shutova et al 2006). Heavy clouds can moderate surface temperatures and decrease solar radiation, both of which would delay GUD (Shutova et al 2006, Richardson et al 2013, Tang et al 2016. The partial correlations between GUD and both precipitation and radiation were opposite in sign (figures 7(c) and (e)), indicating that increased preseason precipitation accompanied by heavy cloud cover may mitigate the impacts of reduced radiation on GUD. In addition, the impacts of drought on GUD changes is also considerable, extreme drought events will cause delaying GUD and the alleviation of drought will cause advancing GUD (Li et al 2023), which is shown by negative correlation between SPEI and GUD in Central Asia and Central east Russia and 6% area with advancing GUD predominated by SPEI changes (figures 7(d) and (3)).

The role of LCC
Different land cover types have different phenological characteristics (Ganguly et al 2010, Jeganathan et al 2014. For example, the GUD of urban vegetation occurs earlier than that of other surrounding vegetation types (Meng et al 2020). Because of the mixed pixel effect in land surface phenology (Chen et al 2018), a coarse grid-such as the 0.05 • grid used in this study-invariably includes several vegetation types. Interannual variation in vegetation composition would change GUD (Peng et al 2017b, Helman 2018, and this change is not caused by climate change. Changes in land cover type also result in changes in surface radiation budget and physiological characteristics that affect the local climate and energy balance (Duveiller et al 2018), thus altering phenology, which have been evidenced by negative correlation between temperature and GUD changes caused by land cover transitions (figure 4 and supplement figure S11). For example, the conversion of forest to shrub in burned area after a forest fire causes GUD to change from a delaying to an advancing trend (Wang and Zhang 2017), which agrees with our results that deforestation of deciduous needleleaf forests advances GUD. The mechanism for this change is the rapid increase in surface albedo and concomitant decrease in evaporation that usually results from the transition from boreal forest to shrub or grassland, which eventually manifests as a drop in temperature in the middle and high latitudes of the Northern Hemisphere (Duveiller et al 2018), thus delaying GUD. Moreover, the temperature of urban areas is higher than that of surrounding rural areas, causing GUD to occur earlier in urban vegetation (Li et al 2017, Meng et al 2020, Tian et al 2020. Therefore, urban expansion usually leads to advancing GUD ( figure 5(c)).

Possible future research
Nitrogen deposition impacts vegetation growth by affecting soil fertility, soil acid-base balance, and nitrogen concentration in leaves (Pan et al 2009, Wu et al 2014. Studies have shown an advanced budding time in response to increased nitrogen deposition in Tibet (Xi et al 2015), and nitrogen deposition contributed to 30.5% of interannual variations in the autumn phenology (Guo et al 2021). The indirect effects of budding time and autumn phenology on GUD (Richardson et al 2013, Tang et al 2016, Piao et al 2019a and the effect of nitrogen deposition on constraining plant phenology need further exploration. The atmospheric CO 2 concentration mainly influences vegetation photosynthesis and reproductive phenophases (Shen et al 2022); however, the mechanism of the impact of increasing CO 2 on GUD is still controversial. There are two possible hypotheses: one is that sufficient CO 2 during vegetation growth enhances frost resistance in vegetation the next year, and the other is that increased CO 2 partly alleviates the negative effects of warming on water availability (Piao et al 2019a). In addition to the effects of nitrogen deposition and CO 2 fertilization on GUD, the ecological impact of invariants in land cover are also worth studying as, for example, forest age and changes in crop cultivar also affect GUD Fabian 1999, Rezaei et al 2018).
We found that LCC may overestimate GUD by 3.3 days (even more than 10 days in some regions). Given that increased carbon sinks due to advance of GUD, could mitigate the risk of climate change to a limited extent (Piao et al 2019b;Chen 2021), the question arises as to whether LCC could offset the beneficial effects of GUD advancement. In terms of relevance for society, the findings of this study may be important for land management and policy formulation. Understanding the role of LCC in shaping the growing season can help inform decisions that may impact carbon uptake and ecosystem services.
In conclusion, by analyzing the statistical relationships between GUD and the individual parameters including temperature, precipitation, radiation, and LCC, we found that LCC exerts a non-negligible impact on GUD. More than 6% of the area with significant GUD change is controlled by LCC. The effect of climate change is overestimated by at least 22% (i.e. 3.3 days) from 1992 to 2020 when LCC is ignored. Our results enrich the understanding of how LCC impacts GUD, allowing us to better understand the climate-driven changes in GUD.

Data availability statement
The data that support the findings of this study are available upon reasonable request from the authors.