Dynamic Changes in Dew Amount in Southern Slope of Boluohuoluo Mountain, Middle Tianshan Mountains

Abstract

Dew is an important water source for natural organisms in arid and semi-arid areas, playing a crucial role in maintaining the stability and sustainability of desert ecosystems. Effectively estimating dew quantity and its long-term changes remains a challenge. Based on conventional meteorological observation data, this study used a Random Forest model to estimate the dew quantity in Nilka, the Southern slope of Boluohuoluo Mountain in middle Tianshan Mountains from June to October in 1970–2022 and analyzed its long-term variations using a statistical method. The results revealed that (1) monthly dewfall varied from 0.74 to 3.88 mm. The 53-year average of the total dew amount in October was significantly higher than in other months (2.81 mm), while the lowest was in August (2.02 mm). In addition, the total dew amount in June, July, and September were 2.27 mm, 2.19 mm, and 2.16 mm, respectively. (2) From 1970 to the beginning of the 21st century, there was a slight decrease in dew from June to October and in every month individually, followed by an increase for about 18 years, after which the dew amount decreased again. During 1970 to 2022, the dew amount exhibited a declining trend when considering the June–October period as a whole or for August and October individually. (3) The change in dew amount was primarily affected by the relative humidity. The findings have implications for assessing the effect of climate change on the dew formation, and could be conducive to further maintaining ecological stability and sustainability in dryland regions amidst global warming.

1. Introduction

Dew is an important water source for the persistence and growth of organisms in arid and semi-arid areas, playing a crucial role in maintaining the stability and sustainability of desert ecosystems. Dew provides a supply of fresh water, forming through the direct condensation of moisture from the ambient atmosphere onto cooler ground, which is an important part of the global hydrological cycle [1]. Dew is a proven essential water source for desert ecosystems and can be directly absorbed by plant foliage and stomata to mitigate water stress, particularly in the arid and semi-arid regions where there is lack of water resources [2]. Thus, the cumulative effect of dew plays an irreplaceable role in maintaining the survival of life and ecological balance in this fragile environment. Dew contributes to water balance by mitigating evaporation-induced water loss in terrestrial ecosystems [3,4,5]. In the arid and semi-arid regions, the amount of dewfall could account for more than 50% of rainfall [5]. Moreover, dew could alleviate vegetation water deficit and promote plant growth [6,7,8,9,10]. In extremely water-scarce desert environments, dew provides essential moisture for the survival of shallow-rooted plants, lichens, microorganisms, and small animals, playing a critical role in maintaining the stability and sustainability of the fragile ecosystem [9,10]. Many desert plants can efficiently absorb dew at night, significantly alleviating water stress during the day and maintaining basic physiological activities for survival and reproduction [6,9]. Therefore, the important role of dew on the sustainability of hydrology and ecosystems cannot be ignored, especially in arid and semi-arid regions [5].

Measuring dew amount is still a challenging task because there is no internationally recognized norm for dew measurement approaches or instruments [11,12]. Since the 1900s, diverse mechanisms for dew measurements have emerged. Among these, artificially engineered surface condensation techniques—including the cloth plate method (CPM) along with plywood, glass, and polyethylene plates—have been established as standardized protocols for quantifying dew formation [2,13,14]. Another approach is to use microcondensators for obtaining the amount of dew by measuring the changes in volume or mass at the onset and cessation of vapor-to-liquid transition [12,15,16,17]. However, these two methods necessitate human-dependent observations on a daily basis throughout the observational time frame, so it is impracticable to procure continuous, long-term observations of dew. Advanced micrometeorological monitoring methods based on energy balance principles, including the eddy covariance system (EC) and Bowen ratio–energy balance method (BREB), have been employed to refine dew quantification frameworks in water-limited ecosystems [18]. They can automatically perform meteorological scans of surface boundary layer meteorological variables and energy fluxes, and this solves the difficulties of continuous observation with the advent of these advanced technologies. However, historical continuous dew data remained unavailable until the advent of these observation systems.

Hybrid semi-analytical and mechanistic modeling frameworks have been established as alternative methodologies to quantify dewfall dynamics in moisture-limited ecosystems. Following the principles of energy balance established by earlier studies, Beysens [19] developed a model for predicting potential dew formation using a limited set of meteorological data. Atashi et al. [20] also utilized an energy balance model to evaluate dew formation potential. Others used conventional the Penman–Monteith (PM) model to simulate nocturnal moisture deposition and used the negative latent heat flux of the model to represent dew [3]. Those results showed that the semi-empirical model and physical model provide relatively accurate estimates of dew [21,22]. However, these models require high temporal resolution data as input. Such data are usually at half-hourly or hourly scales, because dew usually forms at night. These methods are difficult to use when limited to daily data.

Statistical models or machine learning models are well-suited for closing this gap. Hao et al. [11] used 24 h cumulative net radiation along with actual vapor pressure to predict daily dew point amount and duration by multivariate linear regression model (MLR). Lekouch et al. [23] proposed a new predictive framework utilizing artificial neural networks (ANN) to predict dew amounts. They evaluated the potential for dew in 15 major Moroccan cities by applying this model. While many studies have used models like MLR and ANN, recent research highlights that Random Forest (RF) has been validated as a reliable algorithm for estimating ecological hydrological variables [24,25,26], which will help to reconstruct long-term dew formation using meteorological variables combined with RF models.

A large number of condensation observations and forecasts have been made in arid and semi-arid regions based on the above methods. Such studies show that near-surface meteorological parameters are related to those factors which have significant function in the formation of dew [12]. Temperature and humidity levels serve as critical variables governing dew generation [12,27]. Studies show that lower air and surface temperatures, higher relative humidity, and mild wind speeds are conducive to dew formation [28,29]. Evidence from previous research suggests that the Random Forest (RF) method is more precise in simulating and predicting dew amount than other statistical or machine learning models [30]. For this reason, the RF method is employed in the present study.

China’s arid and semi-arid regions, which cover 53 percent of the land area, are expanding every year [12]. Areas with limited water resources in China are key targets of dew-related research. Li et al. [31] conducted a decade-long simulative study in the Tengger Desert, demonstrating that moss-dominated biological soil crusts (biocrusts) significantly enhance dew entrapment compared to bare sandy surfaces, though warming and reduced precipitation may diminish this capacity. Hao et al. [11] studied the formation and long-term trend of dew in the Taklamakan Desert and investigated the amount and duration of dew accumulation in the Desert–Oasis transition zone and valleys [12,30,31]. While Northwest China encompasses both arid and semi-arid ecosystems, existing dew studies are disproportionately concentrated in extreme desert environments [12,31,32]. Therefore, there are few studies on the estimation of dew in the semi-arid region of the middle Tianshan Mountains.

The Tianshan Mountains (TS), the largest mountain system of Central Asia (65°–96° E, 37°–46° N), are located within the arid zone, where westerly circulation brings abundant air moisture. This abundant air moisture plus glacier/snow-melt water generates large proportions of fresh water sources for Central Asia. The TS has an average elevation of roughly 4000 m above sea level. Due to the dominance of an arid continental climate in the TS, a distinct temperature gradient and uneven precipitation distribution are observed. These factors lead to significant differences in climate conditions for dew formation. Nilka is situated on the southern slope of Boluohuoluo Mountain in the middle section of Tianshan Mountains. The large temperature differences between day and night are beneficial to the formation of dew. This study not only addresses a critical knowledge gap regarding long-term dew patterns in this region but, more significantly, provides a novel and representative mountainous case study elucidating the response mechanisms of dew to climate change in global semi-arid regions over a half-century time scale.

Therefore, the main objectives of this study are (1) to reconstruct the dew amount in Nilka by training a Random Forest (RF) model on field-based dew data and meteorological drivers; (2) to analyze the multidecadal dynamics of the reconstructed dew amount and the meteorological variables in Nilka. The findings from this study are critical for elucidating the mechanisms underlying the formation of multidecadal dew patterns in Nilka while providing a framework to analyze the impact of climate perturbations on water cycle dynamics in Nilka.

2. Materials and Methods

2.1. Study Area

This study designated the Nilka region, situated on the Southern slope of Boluohuoluo Mountain in the Middle Tianshan Mountains, as the study area. The terrain slopes downward from the high elevations in the east and north to lower ones in the west and south form a V-shape that opens to the west (Figure 1). The Nilka region extends 243 km in the E–W (east–west) axis and 70 km in the N–S (north–south) dimension [33]. The region belongs to a typical temperate continental climate, characterized by a significant diurnal temperature range and a brief frost-free period. This is mainly affected by the airflow of the Atlantic and Arctic Oceans [34,35]. The altitude in the region is between 800 and 4600 m above sea level. The climate of the region is characterized by an average annual temperature of 8.5 °C. With a total annual precipitation of 350 mm, the region is generally dry, though mountainous areas receive more than 550 mm [36,37]. The main vegetation type is grassland accounting for 65.31% [36], and the main soil type is calcareous soil [33].

2.2. Data

Dew observation data and meteorological data for model training used in this study are all from Yili Station for Watershed Ecosystem Research of Chinese Academy of Sciences (hereinafter referred to as “Yili Station”) (Figure 1), which is located in the Nalati Mountain of the middle Tianshan Mountains with an elevation of about 1400 m, 43°33′06″ N and 84°00′28″ E. According to Feng et al. [30], this study selected the minimum temperature (Tmin), relative humidity (RH), and average wind speed (WS) as predictive factors for model training. Since Nilka Meteorological Station and Yili station are situated at similar latitudes with different longitudes and altitudes, the trained model from Yili station is used to estimate the dew amount in Nilka station.

The meteorological data used to estimate long-term dew amount in this study was obtained from the Nilka National Meteorological Station (hereinafter referred to as “Nilka Station”) with coordinates of 43°48′ N, 82°31′ E and 1106 m above sea level (Figure 1a,d). Considering that the observation period for dew amount was in summer and autumn, we chose a time scale of nearly five months (June to October each year) to estimate the dew amount from 1970 to 2022. Since the data from the Nilka Station only covered the period from 1991 to 2015 and had some data missing, we used ERA5-Land data to interpolate the meteorological station data.

The ERA-Interim dataset constitutes the fourth-generation reanalysis product developed by the European Center for Medium-Range Weather Forecasts (ECMWF), having demonstrated strong performance in Central Asian climate studies. As its successor, the fifth-generation ERA5 reanalysis data exhibit enhanced spatial and temporal resolution alongside improved parameterization methods. This upgraded dataset has proven effective in supporting hydrological simulations and regional climate investigations through systematic validation against observational data. In the ERA5 data, since the 2 m wind speed was not provided, we first derived the 10 m wind speed from the 10 m u- and v-wind components. The 10 m wind speed was then converted to the 2 m wind speed according to the following equation from Allen et al. [38]:

$$\begin{array}{c}{\mathrm{U}}_2= {\mathrm{U}}_{\mathrm{z}}{\displaystyle \frac{4.87}{\ln \left(67.8\mathrm{z}-5.42\right)}}\end{array}$$

(1)

where U₂ is the wind speed at 2 m above the surface (m s⁻¹); Uz is the wind speed (m s⁻¹) at z m above the surface; z is the height (m) of U_z from the surface (in this study, z is 10 m).

ERA5-Land data has demonstrated good performance in representing land surface variables across Xinjiang [39]. In addition, we also compared the ERA5-Land data with weather station data (Figure 2). The correlation coefficients for the reanalysis TA (air temperature), WS, and RH compared to the observed data exceeded 0.87, 0.54, and 0.62, respectively (Figure 2a). A statistical significance (p < 0.05) was found for the correlation coefficients across all reanalysis data when compared to the corresponding observed data. This means that the reanalysis data could explain a significant portion of the observed data’s temporal variation. Therefore, the temporal variations in reanalysis TA, WS, and RH were consistent with the observed ones.

2.3. Random Forest Model

By using bootstrapping, a Random Forest (RF) model derives multiple samples with replacement from the original dataset. At each node, a subset of k attributes is randomly selected, and the best splitting attribute is chosen to construct a decision tree [40]. The Bagging algorithm was used to train n decision trees, and the final result was obtained by voting on the modeling results of all decision trees. This study used the scikit learn library in Python 3.8 to build the RF model. According to Feng et al. [30], the RF model was trained using data from Yili Station, then applied by inputting Tmin, RH, and WS data from Nilka Station to estimate dew amount in Nilka during 1970–2022.

2.4. Analysis of the Trend of Estimated Dew Amount

This research adopted a dual-methodological framework to examine temporal variation patterns in the estimated dew amount in Nilka from 1970 to 2022. The Sen’s slope (β) was used to determine the magnitude of the change in dew amount. The sign of β determines the trend direction: positive for an increase and negative for a decrease. To address the absence of statistical significance testing in this method, the Mann–Kendall (MK) test was applied to analyze the significance of the observed trends [41,42]. According to the MK test statistic Z, if |Z| > 1.96, the trend is significant at the 5% level. According to β and |Z|, we divide the changes in dew amount into five categories: significant increase (β > 0, |Z| > 1.96), significant decrease (β < 0, |Z| > 1.96), increase (β > 0, |Z| ≤ 1.96), decrease (β < 0, |Z| ≤ 1.96), and no significant change [5,30].

3. Results and Analysis

The absence of long-term, direct observations makes it difficult to quantify multidecadal trends in dew amounts. Utilizing ground-based meteorological measurements, this study used RF model to evaluate dew amount. To avoid the errors caused by precipitation, those days which have precipitation have been removed from the statistics and results. The data from the Yili Station was split into a 70% training set and a 30% validation set for model training and validation, respectively. The model’s performance was evaluated using two key metrics: Mean Absolute Error (MAE) of 0.024 mm/d and Root Mean Squared Error (RMSE) of 0.031 mm/d [31]. The model output shows feature importance values of 13.39% for Tmin, 62.76% for RH, and 23.85% for WS.

3.1. Monthly Dew Amount Statistics

Figure 3 shows the monthly average, maximum, minimum, and total monthly dew amount. The data was simulated by the Random Forest model for the period June to October each year from 1970 to 2022, covering a total of 265 months and 8109 days. The minimum simulated monthly dew amount during the simulation period was 0.74 mm (September 2021), and the maximum was 3.88 mm (October 2000) (The bar chart is shown in Figure 3). In terms of daily dew amount, the average was 0.09 mm. Referring to the monthly distribution of daily dew thresholds (Figure 3), the average daily dew amount curve was closer to the minimum daily dew amount, which means the simulation of dew amounts was concentrated at low values.

According to Figure 4, no statistically significant variations were observed in the distribution of monthly total dew amounts among June, July, and September over a 53-year period. August and October demonstrated distinct patterns, with October exhibiting markedly elevated total dew accumulation (the average of 53 years was 2.81 mm) compared to other months, and the total dew amount in August was substantially less than that of other months (the average of 53 years was 2.02 mm). The average total dew amounts for June, July, and September over the 53-year period were 2.27 mm, 2.19 mm, and 2.16 mm, respectively. The mean and median are shown in the boxes as dots and horizontal lines, respectively.

3.2. Long Term Changes in Dew Amount from June to October

The study period from 1970 to 2022 was divided into three stages based on the years of maximum dew: stage 1 (1970–1986), stage 2 (1986–2004), and stage 3 (2004–2020). The trend of dew amount was analyzed in each stage. According to

Table 1. Sen’s slope (β) and M-K test results of dew amount in each month from 1970 to 2022 and from June to October (|Z|) and changing trends.

Month	Temporal Stage	|Z|	β	Trend
June–October	1970–2022	2.67	0	No significant trend
	1970–1986	1.06	−0.01	Decrease
	1986–2004	1.43	0	No significant trend
	2004–2022	3.67	−0.01	Significantly decrease
June	1970–2022	1.89	−0.01	Decrease
	1979–1986	0.37	−0.05	Decrease
	1986–2004	0.42	0.01	Increase
	2004–2022	1.54	−0.02	Decrease
July	1970–2022	1.43	−0.01	Decrease
	1979–1986	0	0	No significant trend
	1986–2004	1.12	0.01	Increase
	2004–2022	1.19	−0.01	Decrease
August	1970–2022	0.3	0	No significant trend
	1979–1986	0.37	−0.02	Decrease
	1986–2004	1.68	0.03	Increase
	2004–2022	0.7	−0.01	Decrease
September	1970–2022	2.48	−0.01	Significantly decrease
	1979–1986	1.11	−0.04	Decrease
	1986–2004	0.14	0	No significant trend
	2004–2022	2.24	−0.05	Significantly decrease
October	1970–2022	0.04	0	No significant trend
	1979–1986	0.87	−0.08	Decrease
	1986–2004	0.35	0	No significant trend
	2004–2022	3.15	−0.08	Significantly decrease

and Figure 5, except for the dew amount in July, which exhibited no statistically significant trend in stage 1, the dew amount in June–October, June, August, September, and October showed a decreasing trend in stage 1. In stage 2, except for the decreasing trend in September, all the remaining months showed an upward trend, but there was no significant trend. In stage 3, a downward trend occurred from June to October, with significant trends in both September and October. A decline was evident in the total dew amount from June to October and for each month when comparing stage 1 to stage 2. However, except for September, the other months showed an increasing trend during this period. The total dew amount then began to decline again in stage 3, with a significant decrease in September and October.

4. Discussion

4.1. Historical Dew Amount

This study used a trained RF model to reconstruct the dew amount in Nilka for the period from June to October of each year, from 1970 to 2022. During these 53 years, the most notable period was from 2004 to 2013, during which the dew amount experienced a nearly 10-year decline. Feng et al. [30] also used RF model to reconstruct the dew amount from 1980 to 2021 in Xinyuan County, a location in the Ili River Valley. They also observed a nearly 10-year decline period in the region’s dew amount (2002–2013). In addition, Hao et al. [11] employed a linear regression algorithm to reconstruct the multidecadal dew amount in the Taklamakan Desert and reported a significant decrease in dew during this period. According to Feng et al. [30], a rapid rise in air temperature and an insignificant change in precipitation resulted in a reduction in relative humidity, which in turn led to a sharp decline in dew amount in the Kunes Valley from 2004 to 2013. During this period, the increase in air temperature in the Kunes Valley was approximately 0.1 °C per year [30]. Therefore, the next section of this study will also analyze the long-term changes in air temperature, relative humidity, and precipitation in Nilka, in an attempt to identify the causes of the dew amount decline during this period.

4.2. Trend of Model Input Variables

Figure 6 shows the trends of the model input variables Tmin, RH, and WS, and precipitation from 1970 to 2022 (where Tmin, RH, and WS are the average values from June to October, and precipitation was the total precipitation from June to October). The years 1986 and 2004 were used as boundaries for the analysis of Tmin, RH, WS, and precipitation. Between 1986 and 2004, there were similar trends in RH, precipitation, and dew. As shown in Figure 6, the RH decreased significantly from 2004 to 2022, indicating that the water vapor in the air was difficult to saturate, which was the direct cause of the decrease in dew. The decrease in RH may also be due to the increase in Tmin and the decrease in precipitation. Figure 6 shows that the Tmin has been rising, with higher temperatures implying that the atmosphere can store more water vapor [38]. At the same time, between 2004 and 2022, precipitation showed a downward trend, indicating a decrease in the sources of atmospheric humidity. Therefore, under the influence of increasing temperature and decreasing precipitation, the amount of dew decreased. As for WS, although the average WS showed a decreasing trend in two phases between 1970 and 2022, the average WS did not exceed 3.5 m/s. According to earlier studies, WS was not conducive to dew production when it was too high, so in this study, WS was not the cause of the decrease in dew.

The observed significant drop in dew during September, which coincides with the fruiting phase of local herbs, suggests a potential link between dew availability and plant reproductive cycles. This temporal pattern implies that dew shortages could exacerbate moisture stress and negatively impact grassland productivity, especially given the Nilka region’s heavy reliance on grassland-dominated landscapes (65.31% cover). Crucially, the observed reduction in dew is intrinsically linked to regional warming and drought trends. Under a global warming scenario, the synergistic effect of rising temperatures and decreasing precipitation may intensify hydrological constraints on ecosystem function within the semi-arid regions of the Tianshan Mountains. Consequently, our results underscore the necessity of integrating dew conservation into regional ecological restoration strategies. Such an approach is vital for mitigating climate-induced water scarcity and ensuring the long-term resilience of these vulnerable ecosystems.

5. Conclusions

Due to the limitations of evaluation methods and the absence of sustained in situ measurements of dew amounts, researchers found it difficult to quantify multidecadal dynamics in dew amount changes. Due to these same limitations, researchers found it difficult to assess the long-term trend of dew amount changes. In order to acquire long-term dew amount data from Nilka, this study employed RF model trained on observational datasets, utilizing meteorological variables as inputs, to reconstruct daily dew amounts from June to October annually between 1970 and 2022. The results showed that the average daily dew was 0.09 mm, the simulated daily dew was concentrated in the low value range, and the monthly dew amount was between 0.74 mm and 3.88 mm. The average daily dew amount in autumn was slightly higher than in summer. Based on the reconstructed values of dew from June to October from 1970 to 2022, the long-term variation in dew was divided into three periods. Between 1970 and 1986, dew decreased from June to October, and between 1986 and 2004 there was a slight increase in dew from June to October, and then it began to experience a period of decline of about 18 years from 2004. The long-term variation in dew was mainly affected by the change in RH. However, this study still has some limitations. The main issue lies in the inability of the Random Forest model to capture the physical mechanisms of dew formation. In the future, machine learning algorithms should be combined with physical process models to estimate dew amount. Additionally, increasing field observations of dew will provide more data for model development and validation.

This study has great significance in exploring the characteristics of historical dew levels and understanding how climate change influences dew formation and supports nature’s balance in arid and semi-arid areas under a changing climate.