Multiscale phosphorus loss in farmland driven by precipitation: effects of farmland type Authors: Kai Shi, Hui Zhang, Xiaoli Mei, Boyun Liu, Qiang Xiao Published: 2026 ABSTRACT Precipitation is the primary driver of phosphorus loss from farmland. However, the multitemporal and spatial characteristics, as well as the mechanisms of phosphorus loss across different farmland types at the watershed scale, remain poorly understood. This study examines the multiscale impacts of different farmland types on phosphorus loss processes driven by long-term precipitation, focusing on two adjacent small watersheds, the Yulin River and the Dahong River, in the Sichuan Basin. First, based on remote sensing data from 2021 to 2022 (precipitation, temperature, land cover) and river water quality monitoring data (total phosphorus (TP), dissolved oxygen (DO), and permanganate index (COD)), the spatiotemporal variations in river water quality are analyzed. The results indicate that under similar climate conditions and pollution source distributions, the downstream TP concentration in the Dahong River (dominated by paddy fields) is significantly lower than that in the upstream, whereas in the Yulin River (dominated by dryland farming), the downstream TP concentration is higher than that in the upstream. Secondly, the study develops a multisource coupling analysis framework that integrates detrended cross-correlation analysis (DCCA), multifractal detrended cross-correlation analysis (MFDCCA), and remote sensing data to evaluate the impact of different farmland types on phosphorus loss behavior. The DCCA analysis results show that precipitation and TP exhibit a clear long-term correlation, with scaling exponents all exceeding 0.5. In the paddy field control area, precipitation and TP exhibit a positive correlation over long time scales. In contrast, in the dryland control area, a positive correlation is observed only over short time scales (< 60 days), while at longer time scales, the correlation turns negative and shows significant fluctuations. MFDCCA further reveals that the coupling relationship between TP and precipitation generally exhibits multifractal characteristics. The multifractal intensity is higher in the dryland-controlled fields (Δh = 1.47), indicating that TP is more sensitive to precipitation perturbations and less stable. The multifractal intensity is lower in the paddy fields (Δh = 1.20), indicating a more stable coupling relationship. Finally, by combining sliding window analysis with 3D convex hull volume calculations, the study quantitatively assesses the distribution differences of multifractal parameters. The results show that areas dominated by paddy fields exhibited stronger aggregation of multifractal parameters (convex hull volume difference of 0.69). In contrast, dryland areas showed more dispersed patterns (convex hull volume difference of –0.32). This study innovatively integrates fractal theory, remote sensing, and watershed observation data to establish a coupling analysis framework for identifying nonpoint source pollution. It elucidates the multiscale response patterns of phosphorus loss under different farmland types. It provides a new technical approach for quantitatively assessing the effects of farmland type on the spatial distribution and migration pathways of TP. Introduction Phosphate fertilizer is an essential nutrient for promoting high crop yields. However, global assessments indicate that phosphorus use efficiency (PUE) in agricultural systems typically ranges between 15% and 30%, with considerable spatial variability (Liu et al., 2025; Zhang et al., 2025). Excessive application can easily lead to phosphorus loss through runoff, leaching, and other pathways. Phosphorus loss not only diminishes soil fertility but also causes eutrophication of water bodies, leading to degradation of water quality, frequent algal blooms, disruption of aquatic ecosystems, and results in significant economic damages and health risks (Wang et al., 2014; Zhou et al., 2022). At present, phosphorus loss from agricultural land poses a significant challenge to controlling diffuse pollution and has gained global concern (Comber et al., 2023; Li et al., 2022; Tong et al., 2023; Wang et al., 2022). In farmland ecosystems, phosphorus loss primarily occurs through surface runoff and soil leaching. Among these, surface runoff is regarded as one of the key factors driving agricultural diffuse pollution and nutrient enrichment in aquatic ecosystems (Wang et al., 2023; Yin et al., 2023). As a primary driver of surface runoff generation, precipitation enhances soil erosion and increases runoff intensity, thereby promoting the transport and export of phosphorus from farmland to river systems (Hou et al., 2022). Especially during high-intensity or prolonged precipitation events, soil moisture increases substantially, leading to increased surface runoff that carries more particulate and dissolved phosphorus (Jin et al., 2020). However, the transport of phosphorus does not exhibit a simple linear response to precipitation but rather shows significant nonlinear characteristics. For example, Li et al. (2021) pointed out that during a rainfall event with a daily rainfall of 33.6 mm and a maximum rainfall intensity of 32.5 mm/h, the total phosphorus (TP) loss was 1.30 kg/ha. In another event with a daily rainfall of 46.0 mm but a maximum rainfall intensity of only 2.92 mm/h, the TP loss increased to 2.93 kg/ha. This nonlinear response mechanism is likely closely related to the soil’s retention and release behavior of phosphorus. In addition, soils of different types and pollution levels exhibit significant differences in their phosphorus retention and release behaviors, resulting in varying phosphorus output responses under similar precipitation conditions (Yang et al., 2024). For instance, Zhao et al. (2023) demonstrated via meta-analysis that phosphorus loss in dryland systems exhibits a greater response to precipitation variability than that in paddy fields across various fertilization conditions. McDowell et al. (2020) simulated and analyzed the time required for WEP (an indicator reflecting phosphorus loss in surface runoff) in phosphorus-enriched soils to decrease to the environmental safety threshold (≤ 0.02 mg/L). The results indicate that the time required for contaminated soil to achieve this goal is generally 36–55 years. For volcanic rock soils with more severe phosphorus accumulation, this process may take over 100 years. In summary, the natural retention of phosphorus by soil and the instability of precipitation itself make the migration and release of phosphorus driven by precipitation show significant time-scale dependence and long-term residual characteristics. This feature is consistent with Stark and Stieglitz’s (2000) description that the apparent migration time of inert solutes exhibits a broad-tailed distribution. The migration process of phosphorus in soil also shows a broad-tailed distribution behavior with multiscale control, multi-stage retention, and slow release. This means that the phosphorus loss caused by the current precipitation is not only directly affected by the precipitation event at that time, but is also deeply affected by historical precipitation, soil accumulation status, and long-term hydrological processes, forming an output characteristic of the precipitation input signal after soil “fractal filtering.” In summary, the phosphorus loss process in farmland is a doubly complex event driven by the randomness of external precipitation and the internal fractal retention mechanism of the soil, exhibiting strong time-scale dependence and long-term nonlinear memory effects. Although the above fractal characteristics are commonly present across different farmland types, factors such as soil physical properties, hydrological conditions, and land management practices lead to significant differences in the specific response mechanisms of this process among various farmlands. Taking paddy fields as an example, ridges and water layers form a substantial hydrological “buffer reservoir” (Atere et al., 2018; Wei et al., 2019), which enhances the soil’s ability to retain phosphorus and slows down the migration process of phosphorus. Liu et al. (2021) demonstrated that enhancing the maximum water retention of paddy fields can effectively reduce surface runoff and phosphorus loss, achieving a reduction of more than 27.18%. Even if heavy precipitation causes surface runoff, the deposition and re-release of phosphorus in the underlying soil are controlled by multiple time scales, resulting in a slower rate of its removal by runoff. The phosphorus migration process exhibits a clear “broad tail” characteristic, which is consistent with the fractal model of pollutant migration proposed by Stark and Stieglitz (2000). In contrast, dryland soils have weaker retention capacity and lack the hydrological regulation features found in paddy fields. Once precipitation intensity exceeds the soil infiltration and storage thresholds, surface runoff can rapidly mobilize phosphorus in the shallow soil layer (Roberts & Johnston, 2015; Yahaya et al., 2023). It exhibits a faster and shorter time-scale transport process, with a relatively narrow “tail” of phosphorus migration, and the response is more pulsed. Therefore, although the retention and slow release of phosphorus in soil generally exhibit fractal characteristics, different types of farmland exhibit significantly different fractal features in response to precipitation due to differences in soil hydrological conditions. Specifically, phosphorus loss in paddy fields tends to show long-term and “delayed release” characteristics, whereas drylands display an event-driven pattern of “rapid response and short-term output.” However, the time-scale characteristics of phosphorus loss dynamics driven by precipitation have not been sufficiently studied. Thus, to better explore the spatiotemporal scale features and complex response mechanisms of phosphorus loss processes under different farmland types, it is necessary to introduce appropriate mathematical tools to quantitatively characterize the multiscale nonlinear relationships between precipitation and phosphorus loss. Fractal methods have demonstrated outstanding performance in uncovering multiscale dependencies among complex environmental variables (Li et al., 2024; Liu et al., 2023; Shi et al., 2025). Among these, detrended cross-correlation analysis (DCCA) is an effective tool for assessing long-range correlations between two time series, and its derived DCCA coefficient can accurately estimate the true correlation between sequences without being affected by non-stationarity. Furthermore, multifractal methods can reveal the long-term memory and multiscale dependency structure of environmental variables by extracting deterministic information from complex fluctuations. For example, Shi et al. (2010) applied the multifractal detrended fluctuation analysis (MFDFA) method to comparatively study the time-scale characteristics of pH values at the inflow and outflow of Poyang Lake. The results indicated that the pH series at the lake’s inflow exhibited long-term memory and multifractal scaling properties, and that both monofractal and multifractal parameters could serve as dynamic indicators quantifying Poyang Lake’s acid-neutralizing capacity. In another study, Adarsh et al. (2020) evaluated the runoff-sediment relationships in five watersheds using multifractal detrended cross-correlation analysis (MFDCCA). The findings showed that the cross-correlation strength between runoff and total suspended solids (TSS) was generally higher on an annual scale than on a seasonal scale. Moreover, at longer time scales, the MFDCCA method was able to distinguish differences in the correlation strength between runoff and TSS across different watersheds. The above research indicates that multifractal methods have good applicability in characterizing nonlinear, multiscale, and long-term memory characteristics in water environment systems. However, the application of fractal methods to phosphorus loss remains limited. Phosphorus loss is affected by various elements, including precipitation and soil retention, resulting in a loss process that exhibits significant multiscale dependence and nonlinear response characteristics. Therefore, introducing DCCA and MFDCCA methods can provide scientific tools for investigating the response mechanisms of phosphorus loss to precipitation across different farmland types. Therefore, to exclude other interfering factors, this study selects the Dahong River and Yulin River as research objects, both of which have similar meteorological conditions, pollution patterns, and levels of human activity, but exhibit significant differences in upstream and downstream farmland type distribution. Using meteorological and water quality data from the two rivers for 2021–2022. Firstly, the long-term cross-correlation relationship between precipitation and TP concentration sequence changes in two watersheds is quantitatively analyzed using DCCA cross-correlation numbers. Secondly, the MFDCCA method is used to deeply explore the multi-fractal characteristics between precipitation and TP concentration in the two basins at multiple time scales. Finally, the sliding window approach is applied to show the evolution characteristics of the correlation between precipitation and TP in the basin at different time scales under the control of different agricultural types. The convex hull volume is used to quantitatively characterize the aggregation degree of precipitation and TP multifractal parameters under different farmland control. This study will provide new insights for quantitatively identifying differences in phosphorus loss processes influenced by different farmland types, and offer scientific support for water pollution control and phosphorus loss management in small watersheds. Materials and methods Study area Yulin River originates from Qinggangzhi, Zhoujia Town, Dazhu County, Dazhou City, Sichuan Province. It is 226km long with a watershed area of 3896 km2. The river is located between the Huaying Mountain, Tongluo Mountain, and Mingyue Mountain on the eastern edge of the Sichuan Basin, spanning from 106° 46′ to 107° 26′ E and from 29° 40′ to 30° 43′ N in geographic coordinates. It flows through Dazhou and Guang’an in Sichuan Province, as well as Chongqing Municipality. The Dahong River serves as a major left-bank branch of the Yulin River, stretching 152 km and draining a watershed covering 1427 km2. The two rivers are geographically adjacent and belong to a typical small watershed system. Water quality monitoring stations have been established at both the upstream and downstream sections of the two rivers. The upstream and downstream stations of the Yulin River are Shuanghekou Bridge and Yaotan. The upstream and downstream stations of the Dahong River are Gangjia Bridge and Cuijiayan Village in Lijia Township. Figure 1 shows the geographic locations of the Yulin River, Dahong River, and their water quality monitoring stations. Fig. 1Geographic locations of the Yulin River, Dahong River, and their water quality monitoring stations Geographic locations of the Yulin River, Dahong River, and their water quality monitoring stations The two rivers share similarities in the following aspects. First, the climatic conditions are similar. The Dahong River and Yulin River basins have average annual temperatures of 18.23 °C and 18.24 °C, respectively, and average annual precipitation of 1208 mm and 1205 mm, respectively. The differences are minimal, indicating comparable climate conditions. Second, the proportion of residential areas and population distribution is close. The proportions of upstream and downstream residential areas for the Dahong River are 1.37% and 4.17%, respectively, while for the Yulin River, they are 1.66% and 4.72%, respectively (see Table 1). The populations of the upstream and downstream reaches of the Dahong River are 0.7 and 1.45 million, respectively, while the populations of the Yulin River are 0.71 and 1.48 million. The overall distribution characteristics are the same. Third, infrastructure configuration is consistent. Both basins have sewage treatment plants located upstream and downstream, with similar treatment capacities and operational modes. Fourth, industrial disturbance factors are similar. Both basins lack major industrial facilities, with primary pollution arising from household wastewater and diffuse agricultural sources. Fifth, the distribution of farmland upstream and downstream is also quite close. The proportion of farmland upstream and downstream of the Dahong River is 69.5% and 70%, respectively, while for the Yulin River, it is 63% and 61%. However, despite these similarities in natural and anthropogenic conditions, the TP concentrations upstream and downstream of the two rivers show significant differences. This phenomenon will be further analyzed and discussed in the following sections. Table 1Proportions of different land cover types in the upstream and downstream areas of the Dahong River and Yulin RiverClassification systemDahong riverYulin riverUpstream area (km2)Downstream area (km2)Upstream proportion (%)Downstream proportion (%)Upstream area (km2)Downstream area (km2)Upstream proportion (%)Downstream proportion (%)Dryland511.31392.6958.4171.18911.981938.6852.3590.00Herbaceous cover cropland0.080.080.010.010.150.320.010.01Paddy field97.04158.8811.0928.80180.31214.9810.359.98Open evergreen broadleaved forest0.000.000.000.000.01/0.00/Closed evergreen broadleaved forest173.29100.8619.8018.28486.54911.9127.9342.34Open deciduous broadleaved forest5.563.420.640.6211.4533.970.661.58Open evergreen needle-leaved forest0.030.010.000.000.000.000.000.00Closed evergreen needle-leaved forest70.9481.528.1014.78113.25311.466.5014.46Grassland1.040.710.120.133.508.720.200.41Swamp0.040.130.000.020.040.280.000.01Marsh0.042.370.000.430.060.390.000.02Flooded flat/0.65/0.120.010.050.000.00Impervious surfaces11.9723.011.374.1728.85101.631.664.72Water body4.0014.970.462.715.8615.120.340.70 Table 1Proportions of different land cover types in the upstream and downstream areas of the Dahong River and Yulin RiverClassification systemDahong riverYulin riverUpstream area (km2)Downstream area (km2)Upstream proportion (%)Downstream proportion (%)Upstream area (km2)Downstream area (km2)Upstream proportion (%)Downstream proportion (%)Dryland511.31392.6958.4171.18911.981938.6852.3590.00Herbaceous cover cropland0.080.080.010.010.150.320.010.01Paddy field97.04158.8811.0928.80180.31214.9810.359.98Open evergreen broadleaved forest0.000.000.000.000.01/0.00/Closed evergreen broadleaved forest173.29100.8619.8018.28486.54911.9127.9342.34Open deciduous broadleaved forest5.563.420.640.6211.4533.970.661.58Open evergreen needle-leaved forest0.030.010.000.000.000.000.000.00Closed evergreen needle-leaved forest70.9481.528.1014.78113.25311.466.5014.46Grassland1.040.710.120.133.508.720.200.41Swamp0.040.130.000.020.040.280.000.01Marsh0.042.370.000.430.060.390.000.02Flooded flat/0.65/0.120.010.050.000.00Impervious surfaces11.9723.011.374.1728.85101.631.664.72Water body4.0014.970.462.715.8615.120.340.70 Proportions of different land cover types in the upstream and downstream areas of the Dahong River and Yulin River Data sources This study utilizes water quality data obtained from the Sichuan Province Ecological Environment Monitoring Station. The sampling period for this study is from January 1, 2021, to December 31, 2022. In the water quality data, dissolved oxygen (DO) data is hourly, while TP and COD data are 4-hourly. Due to instrument maintenance and equipment failures, some data were missing at various monitoring sites. Missing values are filled by averaging the concentrations before and after the missing time points. The proportion of missing data does not exceed 5%. To synchronize the analysis with rainfall, the hourly or 4-hourly water quality data are averaged to obtain daily average water quality data. Due to the inability of traditional precipitation monitoring stations to provide accurate data for every region. Therefore, to more accurately capture the impact mechanism of precipitation on total phosphorus in rivers. This study uses the CHM_PRE daily precipitation dataset released by the National Tibetan Plateau Data Center (TPDC) (Miao, 2023) , with a spatial resolution of 0.1°. The dataset contains daily precipitation records from 2839 meteorological stations in China and surrounding areas from 1961 to 2022. Compared to the traditional method of constructing precipitation background and ratio fields, CHM_PRE incorporates monthly precipitation constraints and terrain corrections, and its accuracy has been validated using interpolated daily precipitation data from approximately 40,000 high-density stations during 2015–2019 (Gou et al., 2021; Han et al., 2023; Miao et al., 2022). This dataset has been widely applied in hydrological process modeling and climate change studies (Tang et al., 2024; Wei et al., 2025). In this study, ArcGIS software is used to mask the precipitation dataset according to the study areas, extracting the daily precipitation raster data for the Yulin River and Dahong River basins. The precipitation raster dataset was clipped using watershed boundaries in ArcGIS. All grid cells intersecting each watershed were extracted. Basin-scale precipitation was calculated using an area-weighted average method rather than simple summation to ensure physical consistency and avoid artificial amplification of rainfall magnitude. And the corresponding precipitation time series are constructed. The processing method of temperature data is the same as above. The land cover type data are sourced from the National Glacier and Frozen Soil Desert Science Data Center. Its spatial resolution is 30 m. The distribution of different types of vegetation in the Yulin River and Dahong River basins can be obtained by ArcGIS. Research methods Detrended cross-correlation coefficient The detrended cross-correlation coefficient is a new metric based on the detrended fluctuation analysis (DFA) and detrended cross-correlation analysis (DCCA) methods, which can be used to quantitatively characterize the level of cross-correlation between precipitation and TP. Detailed methodological descriptions can be found in Gvozdanovic et al. (2012) and Qian et al. (2015). Step 1: Take two time series representing precipitation and TP, denoted as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${x}_{t}$$\end{document}xt and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${y}_{t}$$\end{document}yt for\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$t = 1, 2, ..., N$$\end{document}t=1,2,...,N. Then, compute the cumulative sum (integrated series) for each time series.1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$xx_k=\sum_{t=1}^kx_t,yy_k=\sum_{t=1}^ky_t,k=\mathrm{1,2},\dots,N.$$\end{document}xxk=∑t=1kxt,yyk=∑t=1kyt,k=1,2,⋯,N.Step2: Divide the two time series of precipitation and TP, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${x}_{t}$$\end{document}xt and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${y}_{t}$$\end{document}yt into \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$(N-s)$$\end{document}(N-s) equal-length overlapping segments, where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$4 \le s \le \frac{N}{4}$$\end{document}4≤s≤N4.Step3: By performing least squares fitting on each segment \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$x{P}_{i}(k)$$\end{document}xPi(k) and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$y{P}_{i}(k)$$\end{document}yPi(k), estimate the local trend within each segment, then calculate the covariance of the residuals for segment \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$s$$\end{document}s using the following equation:2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$F_{xy}^2\left(s,i\right)=\frac1{N+1}\sum_{k=1}^{i+s}\left[xx_k-xP_i(k)\right]\left[yy_k-yP_i(k)\right]$$\end{document}Fxy2s,i=1N+1∑k=1i+sxxk-xPi(k)yyk-yPi(k) Step2: Divide the two time series of precipitation and TP, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${x}_{t}$$\end{document}xt and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${y}_{t}$$\end{document}yt into \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$(N-s)$$\end{document}(N-s) equal-length overlapping segments, where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$4 \le s \le \frac{N}{4}$$\end{document}4≤s≤N4. Step2: Divide the two time series of precipitation and TP, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${x}_{t}$$\end{document}xt and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${y}_{t}$$\end{document}yt into \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$(N-s)$$\end{document}(N-s) equal-length overlapping segments, where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$4 \le s \le \frac{N}{4}$$\end{document}4≤s≤N4. Step3: By performing least squares fitting on each segment \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$x{P}_{i}(k)$$\end{document}xPi(k) and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$y{P}_{i}(k)$$\end{document}yPi(k), estimate the local trend within each segment, then calculate the covariance of the residuals for segment \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$s$$\end{document}s using the following equation:2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$F_{xy}^2\left(s,i\right)=\frac1{N+1}\sum_{k=1}^{i+s}\left[xx_k-xP_i(k)\right]\left[yy_k-yP_i(k)\right]$$\end{document}Fxy2s,i=1N+1∑k=1i+sxxk-xPi(k)yyk-yPi(k) Step3: By performing least squares fitting on each segment \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$x{P}_{i}(k)$$\end{document}xPi(k) and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$y{P}_{i}(k)$$\end{document}yPi(k), estimate the local trend within each segment, then calculate the covariance of the residuals for segment \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$s$$\end{document}s using the following equation:2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$F_{xy}^2\left(s,i\right)=\frac1{N+1}\sum_{k=1}^{i+s}\left[xx_k-xP_i(k)\right]\left[yy_k-yP_i(k)\right]$$\end{document}Fxy2s,i=1N+1∑k=1i+sxxk-xPi(k)yyk-yPi(k) Perform the above procedure for all \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$(N-s)$$\end{document}(N-s) overlapping segments and compute their average to derive a new covariance function:3\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$F_{xy}^2\left(s\right)=\frac1{(N-s)}\sum_{i=1}^{N-s}f_{xy}^2\left(s,i\right)$$\end{document}Fxy2s=1(N-s)∑i=1N-sfxy2s,i If a long-range cross-correlation exists between precipitation and TP, it follows a power-law form, namely,4\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$F_{xy}\left(s\right)\propto s^\lambda$$\end{document}Fxys∝sλwhere \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\lambda$$\end{document}λ is the DCCA exponent, which reflects the long-term cross-correlation for precipitation and TP. When \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\lambda \approx 0.5$$\end{document}λ≈0.5, it indicates that there is no long-term cross-correlation between precipitation and TP, meaning that the two variables are completely independent at any given time. When \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\lambda \ne 0.5$$\end{document}λ≠0.5, it suggests the presence of long-term cross-correlation, implying that the values of precipitation and TP are not entirely independent. A value of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\lambda > 0.5$$\end{document}λ>0.5 denotes a persistent long-term positive correlation, meaning that an increasing (or decreasing) trend in precipitation over a past period will lead to a similar trend in TP, with this influence exhibiting a scale-free power-law decay over certain time scales. Conversely, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\lambda < 0.5$$\end{document}λ<0.5 indicates an anti-persistent long-term cross-correlation between precipitation and TP. Since the DCCA exponent can only indicate whether a long-range power-law cross-correlation exists between precipitation and TP, but cannot quantify its strength, Zebende (2011) proposed the DCCA cross-correlation coefficient to measure the degree of cross-correlation between nonstationary time series. This study uses this coefficient to quantitatively characterize the relationship between precipitation and TP. The DCCA cross-correlation coefficient \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\rho }_{DCCA}(s)$$\end{document}ρDCCA(s) can be expressed as:5\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\rho_{DCCA}\left(s\right)=\frac{F_{xy}^2\left(s\right)}{F_{xx}\left(s\right)F_{yy}\left(s\right)}$$\end{document}ρDCCAs=Fxy2sFxxsFyys\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\rho }_{DCCA}(s)$$\end{document}ρDCCA(s) represents a set of cross-correlation coefficients associated with scale \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$s$$\end{document}s, used to describe how the cross-correlation varies across different time scales. One advantage lies in its capability to assess the correlation between two non-stationary time series across multiple time scales. Its values range from \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$-1 \le {\rho }_{DCCA} \le 1$$\end{document}-1≤ρDCCA≤1. A \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\rho }_{DCCA}$$\end{document}ρDCCA value of 1 reflects a complete positive relationship between precipitation and TP; a value of −1 denotes a perfect inverse association;while two variables. Multifractal detrended cross-correlation analysis (MF-DCCA) MF-DCCA is a novel statistical analysis method capable of revealing the interrelationship between precipitation and TP, as well as their multifractal characteristics. The calculation steps are as follows: First, perform cumulative deviation processing on the precipitation series \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left\{x(r)\right\}\left(r=\mathrm{1,2},\cdots ,N\right)$$\end{document}x(r)r=1,2,⋯,N and TP series \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left\{y(r)\right\}(r=\mathrm{1,2},\cdots ,N)$$\end{document}y(r)(r=1,2,⋯,N) from the Dahong River and Yulin River basins to obtain the cumulative deviation sequences \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left\{X(t)\right\}$$\end{document}X(t) and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left\{Y(t)\right\}$$\end{document}Y(t):6\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{array}{c}X\left(t\right)=\sum_{r=1}^N\left[x\left(r\right)-\overline x\right]\end{array}$$\end{document}Xt=∑r=1Nxr-x¯7\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{array}{c}Y\left(t\right)=\sum_{r=1}^N\left[y\left(r\right)-\overline y\right]\end{array}$$\end{document}Yt=∑r=1Nyr-y¯ In the formula, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\overline{x }$$\end{document}x¯ and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\overline{y }$$\end{document}y¯ represent the mean values of the studied series, where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\overline{x }=\frac{1}{N}\sum_{r=1}^{N}x\left(r\right)\text{ and }\overline{y }=\frac{1}{N}\sum_{r=1}^{N}y(r)$$\end{document}x¯=1N∑r=1Nxrandy¯=1N∑r=1Ny(r). Then, the cumulative deviation sequences \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left\{X(t)\right\}$$\end{document}X(t) and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left\{Y(t)\right\}$$\end{document}Y(t) are divided into consecutive non-overlapping segments. The number of segments is \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${N}_{s}=\frac{N}{s}$$\end{document}Ns=Ns, where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$s$$\end{document}s is the length of the time scale. Because \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$N$$\end{document}N might not be evenly divisible by s, and to prevent data loss, the sequences are additionally segmented from the opposite end using the same method, yielding a total of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$2{N}_{s}$$\end{document}2Ns segments. For each small area \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$v$$\end{document}v, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$v=\mathrm{1,2},\cdots ,2{N}_{s}$$\end{document}v=1,2,⋯,2Ns, a linear fit is performed using the least squares method to obtain the local trends \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\widetilde{X(t)}$$\end{document}X(t)~ and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\widetilde{Y(t)}$$\end{document}Y(t)~. Next, the residual covariance within each segment \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$v$$\end{document}v is computed using the following formula:8\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{array}{c}\begin{array}{c}F^2\left(s,v\right)=\frac1s\sum_{t=1}^s\left|X\left(\left(v-1\right)s+t\right)-\widetilde{X\left(t\right)}\right|\bullet\left|Y\left(\left(v-1\right)s+t\right)-\widetilde{Y\left(t\right)}\right|,v=\mathrm{1,2},\cdots,N_s\end{array}\end{array}$$\end{document}F2s,v=1s∑t=1sXv-1s+t-Xt~∙Yv-1s+t-Yt~,v=1,2,⋯,Ns9\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{array}{c}\begin{array}{c}F^2\left(s,v\right)=\frac1s\sum_{t=1}^s\left|X\left(N-\left(v-N_s\right)s+t\right)-\widetilde{X\left(t\right)}\right|\bullet\left|Y\left(N-\left(v-N_s\right)s+t\right)-\widetilde{Y\left(t\right)}\right|\end{array}\end{array}$$\end{document}F2s,v=1s∑t=1sXN-v-Nss+t-Xt~∙YN-v-Nss+t-Yt~ At this point, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$v={N}_{s}+1,{N}_{s}+2,\cdots ,2{N}_{s}$$\end{document}v=Ns+1,Ns+2,⋯,2Ns. Next, the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$q$$\end{document}q order fluctuation function \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${F}_{q}(s)$$\end{document}Fq(s) of the sample (where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$q(q\ne 0)$$\end{document}q(q≠0)) is calculated as:10\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{array}{c}F_q\left(s\right)=\left\{\frac1{2N_s}\sum_{v=1}^{2N_s}\left[F^2\left(s,v\right)\right]^\frac q2\right\}^\frac1qq\neq0\end{array}$$\end{document}Fqs=12Ns∑v=12NsF2s,vq21qq≠011\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{array}{c}F_q\left(s\right)=exp\left\{\frac1{4N_s}\sum_{v=1}^{2N_s}\mathrm{ln}\left[F^2\left(s,v\right)\right]\right\}q=0\end{array}$$\end{document}Fqs=exp14Ns∑v=12NslnF2s,vq=0 In this study, the parameter \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$q$$\end{document}q varies between −20 and 20. Notably, when \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$q$$\end{document}q equals 2, MF-DCCA reduces to the standard DCCA method. Finally, if a long-range power-law cross-correlation exists between the precipitation and TP time series, then \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${F}_{q}(s)\propto {s}^{h(q)}$$\end{document}Fq(s)∝sh(q), where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$h(q)$$\end{document}h(q) denotes the generalized Hurst exponent. Specifically, when \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$q=2$$\end{document}q=2, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$h(q)$$\end{document}h(q) corresponds to the cross-correlation exponent \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\lambda$$\end{document}λ in the DCCA method. A value of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$h(2)=0.5$$\end{document}h(2)=0.5 indicates the absence of correlation between the two sequences. \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$h(2)>0.5$$\end{document}h(2)>0.5 indicates the presence of a long-term power-law cross-correlation between precipitation and TP sequences, suggesting that an increase in precipitation may lead to a rise in river TP over a certain future time scale, meaning that the correlation between precipitation and TP exhibits long-term persistence. Conversely, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$h(2)<0.5$$\end{document}h(2)<0.5 indicates the opposite. If \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$h(q)$$\end{document}h(q) decreases monotonically with increasing \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$q$$\end{document}q, it indicates that the cross-correlation between the sequences displays classic multifractal behavior. The degree of multifractality can be measured by the span of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$h(q)$$\end{document}h(q):12\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{array}{c}\Delta h=h_\mathrm{max}-h_\mathrm{min}\end{array}$$\end{document}Δh=hmax-hminwhere \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta h$$\end{document}Δh represents the strength of the multifractality. The larger the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta h$$\end{document}Δh, the stronger the multifractal nature. \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${h}_{\mathrm{max}}$$\end{document}hmax (usually corresponding to smaller negative orders of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$q$$\end{document}q, such as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$q = -20$$\end{document}q=-20) reflects the correlation strength in regions of small fluctuations within the system, representing the coupling characteristics of the stable and slowly varying parts. In contrast, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${h}_{\mathrm{min}}$$\end{document}hmin (usually corresponding to larger positive orders of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$q$$\end{document}q, such as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$q = 20$$\end{document}q=20) represents the correlation strength in regions of large fluctuations, indicating the coupling characteristics of sudden changes and intense variations within the system. The methodological framework diagram is shown in Fig. 2. Fig. 2Schematic diagram of research methods framework Schematic diagram of research methods framework Results and discussion Analysis of water quality characteristics Figure 3a–i shows the statistical characteristics of precipitation, temperature, and TP concentrations in the upstream and downstream areas of the Dahong River and Yulin River from January 2021 to December 2022. As seen in Fig. 3a–f, the precipitation and temperature trends in both river basins are consistent, indicating similar meteorological conditions with no significant differences. However, there are notable differences between the upstream and downstream areas of the two basins in terms of the spatial distribution and variation trends of TP concentrations. Fig. 3a Precipitation statistics for the Dahong River and Yulin River basins from 2021 to 2022. (b) and (c) monthly average precipitation statistics for the Dahong River and Yulin River basins, respectively. (d) temperature statistics for the Dahong River and Yulin River basins from 2021 to 2022. (e) and (f) monthly average temperature statistics for the Dahong River and Yulin River basins, respectively. (g) TP statistics for the upstream and downstream basins of the Dahong River and Yulin River from 2021 to 2022. (h) and (i) monthly average TP statistics for the upstream and downstream basins of the Dahong River and Yulin River, respectively.) a Precipitation statistics for the Dahong River and Yulin River basins from 2021 to 2022. (b) and (c) monthly average precipitation statistics for the Dahong River and Yulin River basins, respectively. (d) temperature statistics for the Dahong River and Yulin River basins from 2021 to 2022. (e) and (f) monthly average temperature statistics for the Dahong River and Yulin River basins, respectively. (g) TP statistics for the upstream and downstream basins of the Dahong River and Yulin River from 2021 to 2022. (h) and (i) monthly average TP statistics for the upstream and downstream basins of the Dahong River and Yulin River, respectively.) Specifically, the annual average TP concentration in the Yulin River increases from 0.08 mg/L upstream to 0.12 mg/L downstream, showing higher concentrations downstream than upstream. In contrast, the Dahong River exhibits the opposite trend, with its annual average TP concentration decreasing from 0.12 mg/L upstream to 0.05 mg/L downstream (Fig. 3g). From the monthly variation perspective, the TP concentration upstream in the Dahong River is significantly higher than downstream throughout the year. The upstream TP concentration peaks in February, then decreases month by month, reaching its lowest point in winter (Fig. 3h). For the Yulin River, TP concentration downstream is higher than upstream. The downstream TP concentration also peaks in February, followed by a downward trend, reaching its lowest values in August and winter (Fig. 3i). Figure 4 and Table 1 reflect the land cover types in the Dahong River and Yulin River basins. Although the two basins are similar in overall land use structure, vegetation types, and the proportion of impervious surfaces, there are significant differences in the types of farmland downstream. Specifically, dryland accounts for as much as 90.00% downstream in the Yulin River basin, compared to 71.18% downstream in the Dahong River basin. Meanwhile, paddy field and water body proportions downstream in the Dahong River basin are 28.80% and 2.71%, respectively, significantly higher than the 9.98% and 0.70% observed downstream in the Yulin River basin. Fig. 4Land cover types in the Yulin River and Dahong River basins Land cover types in the Yulin River and Dahong River basins The above results indicate that there are significant differences in TP concentrations between the upstream and downstream areas of the Dahong River and Yulin River basins. However, since the two basins have similar meteorological conditions and major anthropogenic pollution sources (such as domestic sewage and industrial wastewater), these factors alone cannot adequately explain the spatial heterogeneity of TP. Instead, differences in farmland types, especially the proportion and spatial distribution of paddy fields versus dryland, may be the key factors driving the spatial variation in water quality. Furthermore, from the monthly variation characteristics of TP, the downstream TP concentration trend in the Dahong River aligns closely with the seasonality of precipitation, suggesting that TP concentrations in this region are significantly influenced by precipitation-driven runoff. The TP concentration peaks observed in February in the upstream Dahong River and downstream Yulin River are likely closely related to spring agricultural activities such as crop sowing and fertilization. After February, TP concentrations decrease and reach their lowest levels in August and winter. The frequent heavy precipitation in August may have a cleansing effect on the rivers, while reduced fertilization in winter leads to more stable river water quality. DCCA cross-correlation coefficient between TP and precipitation To quantitatively analyze the impact of precipitation on the evolution of TP concentration in the river basins, as well as the long-term cross-correlation between the two. This study employed DCCA and the DCCA cross-correlation coefficient to analyze the precipitation and TP relationship. Figure 5a–b show the DCCA results between precipitation and TP in the upstream and downstream regions of the Yulin River and Dahong River. Fig. 5a shows the DCCA analysis of TP and precipitation in the Dahong River. (b) shows the DCCA analysis of TP and precipitation in the Yulin River. (c) shows the DCCA correlation coefficient analysis of TP and precipitation in the Dahong River, and (d) shows the DCCA correlation coefficient analysis of TP and precipitation in the Yulin River a shows the DCCA analysis of TP and precipitation in the Dahong River. (b) shows the DCCA analysis of TP and precipitation in the Yulin River. (c) shows the DCCA correlation coefficient analysis of TP and precipitation in the Dahong River, and (d) shows the DCCA correlation coefficient analysis of TP and precipitation in the Yulin River As shown in Fig. 5a–b, the precipitation and TP relationship in both the upstream and downstream regions of the Dahong River and Yulin River exhibit a strong linear relationship in the logarithmic coordinates of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ln[F\left(s\right)]$$\end{document}ln[Fs] versus \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ln(s)$$\end{document}ln(s). Using the least squares method, the DCCA scaling exponent \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\lambda$$\end{document}λ for the precipitation and TP time series of the upstream and downstream of both rivers is linearly estimated. The DCCA scaling exponents for the Dahong River upstream and downstream are 0.96 and 0.98, respectively, while for the Yulin River upstream and downstream, they are 0.76 and 1.23, respectively. Since all \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\lambda$$\end{document}λ values are greater than 0.5, it indicates that the correlation between precipitation and TP in both upstream and downstream regions of the two rivers exhibits strong long-term persistence. This means that an increase in precipitation over a past period leads to a sustained increase in TP concentration over a future time scale. Figure 5c–d further illustrate the variation of the DCCA cross-correlation coefficient (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\rho }_{DCCA}$$\end{document}ρDCCA) with time scale for the upstream and downstream regions of the two rivers. The figure shows that in the upstream sections of the two rivers, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\rho }_{DCCA}$$\end{document}ρDCCA exhibits a weak positive correlation at small time scales (s < 60 days). It then shifts to a negative correlation at medium scales (60 < s < 360 days), and gradually increases again at larger scales (s > 360 days). In contrast, the downstream areas show markedly different trends. In the downstream section of the Dahong River, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\rho }_{DCCA}$$\end{document}ρDCCA shows a continuous upward trend overall. Except for a brief negative correlation at the very small scale (s < 1 day), the correlations at other scales are significantly positive, reaching up to 0.8 at large scales (s > 365 days). In comparison, the downstream section of the Yulin River exhibits a \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\rho }_{DCCA}$$\end{document}ρDCCA trend similar to its upstream. These results indicate that the influence of precipitation on phosphorus loss is not a consistently positive or negative correlation over long time scales, but rather alternates between positive and negative due to the combined effects of multiple factors, and show different trends in different regions. The explanation for the variation trend of the precipitation and TP \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\rho }_{DCCA}$$\end{document}ρDCCA with time scale is as follows: The downstream region of the Dahong River shows a \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\rho }_{DCCA}$$\end{document}ρDCCA trend significantly different from its upstream area and from both upstream and downstream regions of the Yulin River, which may be related to land use types. This area has a high proportion of paddy fields (28.8%, see Table 1). During precipitation events, at small scales (s < 1 day), the bund structures of the paddy fields function to retain water, causing rainwater to be temporarily held within the fields and suppressing phosphorus loss. This results in a negative correlation between precipitation and TP concentration. However, the retained water creates an anaerobic environment in the paddy fields, accelerating the release of phosphorus from the soil, causing TP concentration to gradually increase. As precipitation continues and exceeds the bund’s water retention capacity, the excess water carries high concentrations of phosphorus out of the fields, leading to a rapid increase in \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\rho }_{DCCA}$$\end{document}ρDCCA after s > 1 day and a shift to a significant positive correlation. In contrast, the upstream area of the Dahong River and both upstream and downstream regions of the Yulin River have a smaller proportion of paddy fields and are mainly dominated by dryland, lacking the water retention and buffering function of paddy fields. When precipitation occurs, surface runoff forms rapidly, carrying phosphorus loss from the topsoil. However, this runoff is still intercepted by vegetation. Therefore, at small scales (s < 60 days), a weak positive correlation is observed. As precipitation continues to increase, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\rho }_{DCCA}$$\end{document}ρDCCA shows a slight upward trend. However, as time progresses (s > 60 days), vegetation grows more vigorously, enhancing its ability to intercept surface runoff and reduce phosphorus loss, which causes \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\rho }_{DCCA}$$\end{document}ρDCCA to shift to a negative correlation. When the time scale further expands beyond 1 year (s > 365 days), vegetation enters a new growth phase, and the soil vegetation system’s regulation of phosphorus changes, leading to a rebound in \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\rho }_{DCCA}$$\end{document}ρDCCA. DCCA cross-correlation coefficient between DO and COD The abundance of phosphorus in the water promotes the growth of algae in rivers. DO and COD are key indicators reflecting algal growth in rivers (Wang et al., 2020). Therefore, to further verify whether the differences in TP and precipitation correlations observed in the downstream area of the Dahong River are influenced by differences in farmland types. This study analyzes DO and COD using DCCA and the DCCA correlation coefficient. The analysis results are shown in Fig. 6a–d. Fig. 6a shows the DCCA analysis of DO and COD in the Dahong River. (b) shows the DCCA analysis of DO and COD in the Yulin River. (c) shows the DCCA cross-correlation coefficient analysis of DO and COD in the Dahong River, and (d) shows the DCCA cross-correlation coefficient analysis of DO and COD in the Yulin River a shows the DCCA analysis of DO and COD in the Dahong River. (b) shows the DCCA analysis of DO and COD in the Yulin River. (c) shows the DCCA cross-correlation coefficient analysis of DO and COD in the Dahong River, and (d) shows the DCCA cross-correlation coefficient analysis of DO and COD in the Yulin River From Fig. 6a–-b, it can be seen that the upstream and downstream of Dahong River and Yulin River exhibit a good linear relationship in the logarithm \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ln[F\left(s\right)]$$\end{document}ln[Fs] and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ln(s)$$\end{document}ln(s), indicating a significant power-law correlation between DO and COD. The DCCA scaling indices of DO and COD in the upstream and downstream of Dahong River and Yulin River are 1.22, 0.99, 1.33, and 1.28, respectively. Since all \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\lambda$$\end{document}λ values are greater than 0.5, it indicates that the correlation between DO and COD in both rivers exhibits strong long-term persistence. This means that an increase in DO (or COD) over a certain past period will lead to a sustained increase in COD (or DO) concentrations over a future period. Figure 6c–d present the DCCA cross-correlation coefficient analysis results between DO and COD in the upstream and downstream regions of the Dahong River and Yulin River. As shown, in the upstream areas of both rivers, DO and COD exhibit a stable positive correlation at small time scales (s < 60 days). At larger time scales (60 days < s < 365 days), the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\rho }_{DCCA}$$\end{document}ρDCCA between COD and DO declines, showing a negative correlation. When the time scale exceeds 1 year (s > 365 days), the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\rho }_{DCCA}$$\end{document}ρDCCA between COD and DO gradually rises again. The trend of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\rho }_{DCCA}$$\end{document}ρDCCA for COD and DO in the downstream regions of both rivers is consistent with that of their upstream areas. However, it is noteworthy that although the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\rho }_{DCCA}$$\end{document}ρDCCA trend downstream in the Dahong River aligns with its upstream, the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\rho }_{DCCA}$$\end{document}ρDCCA remains positively correlated across the entire time scale. These results indicate that the cross-correlation relationship between COD and DO over long time scales is not simply constant but alternates between positive and negative correlations under the influence of multiple factors. The changing pattern of COD and DO \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\rho }_{DCCA}$$\end{document}ρDCCA with respect to time scale is interpreted as follows: In the downstream area of the Dahong River, phosphorus accumulated in the paddy fields continuously flows into the river during precipitation events, causing frequent algal blooms and resulting in an overall positive correlation between DO and COD. However, it is worth noting that after about 2 months (60 < s < 365 days), the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\rho }_{DCCA}$$\end{document}ρDCCA downstream in the Dahong River declines to some extent, likely related to the massive death and decomposition following the concentrated algal bloom. This process significantly consumes dissolved oxygen, causing DO to decrease and COD to further increase, thereby weakening the positive correlation. Subsequently, at larger time scales (s > 365 days), as algae begin a new growth cycle, the correlation between DO and COD strengthens again. In contrast, the upstream area of the Dahong River and both upstream and downstream areas of the Yulin River are dominated by dryland, where phosphorus loss occurs intermittently. During spring planting, phosphorus loss with precipitation runoff is relatively high, while during vigorous vegetation growth, phosphorus loss is relatively low. Therefore, the COD and DO \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\rho }_{DCCA}$$\end{document}ρDCCA exhibits a positive correlation at small time scales (s < 60 days), switches to negative correlation at medium time scales (60 < s < 365 days) due to algal decay, and gradually rises again after the annual scale (s > 365 days). TP and precipitation MFDCCA Multifractal characteristics Previous studies based on DCCA analysis revealed significant differences in the long-term cross-correlation between TP and precipitation under different farmland types, suggesting that their relationship may possess a more complex multiscale structure. Therefore, it is necessary to further analyze TP and precipitation using the MFDCCA method to characterize their spatiotemporal heterogeneity across different time scales. Figures 7 and 8 show the MFDCCA analysis results of TP and precipitation in the upstream and downstream of the Yulin River and Dahong River. As shown in Figs. 7a–b and 8a–b, from 2021 to 2022, there is a good linear relationship between TP precipitation \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$In(s)-InFq(s)$$\end{document}In(s)-InFq(s) in the upstream and downstream of Dahong River and Yulin River on a large time scale (only fitting lines with q = 0, ± 5, ± 10, and ± 20 are shown in the figures). This indicates that the correlation between TP and precipitation follows a power-law distribution. In other words, at a given time scale, the variation in the correlation between TP and precipitation does not follow an exponentially decaying Markov process, but instead decays slowly in a power-law form. Therefore, the scale characteristics of the correlation between TP and precipitation exhibit multifractal scale-invariance properties. Fig. 7MFDCCA analysis results of TP and Precipitation in the Yulin River. (a) and (b) show the double-logarithmic fluctuation curves of TP and precipitation in the upstream and downstream 633 sections of the Yulin River, respectively. (c) and (d) show the q-order generalized Hurst exponents 634 of TP and precipitation in the upstream and downstream sections of the Yulin River, respectivelyFig. 8MFDCCA analysis results of TP and Precipitation in the Dahong River. (a) and (b) show the double-logarithmic fluctuation curves of TP and precipitation in the upstream and downstream sections of the Dahong River, respectively. (c) and d show the q-order generalized Hurst exponents of TP and precipitation in the upstream and downstream sections of the Dahong River, respectively MFDCCA analysis results of TP and Precipitation in the Yulin River. (a) and (b) show the double-logarithmic fluctuation curves of TP and precipitation in the upstream and downstream 633 sections of the Yulin River, respectively. (c) and (d) show the q-order generalized Hurst exponents 634 of TP and precipitation in the upstream and downstream sections of the Yulin River, respectively MFDCCA analysis results of TP and Precipitation in the Dahong River. (a) and (b) show the double-logarithmic fluctuation curves of TP and precipitation in the upstream and downstream sections of the Dahong River, respectively. (c) and d show the q-order generalized Hurst exponents of TP and precipitation in the upstream and downstream sections of the Dahong River, respectively Further analysis shows that \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$h(q)$$\end{document}h(q) exhibits a monotonically decreasing trend as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$q$$\end{document}q varies from −20 to 20 (the blue lines in Figs. 7c–d and 8c–d), rather than remaining constant, indicating that the cross-correlation between precipitation and TP possesses significant multifractal characteristics. This feature reflects the heterogeneity in the strength of their correlation across different time scales. The multifractal intensity can be quantitatively characterized by the multifractal parameter \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta h$$\end{document}Δh (defined as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta h = {h}_{\mathrm{max}} - {h}_{\mathrm{min}}$$\end{document}Δh=hmax-hmin). A larger \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${h}_{\mathrm{max}}$$\end{document}hmax indicates that under normal fluctuation conditions, the coupling relationship between precipitation and TP is more significant, reflecting strong system linkage and better water quality adaptability to precipitation changes. A larger \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${h}_{\mathrm{min}}$$\end{document}hmin suggests that even under strong disturbances or extreme precipitation events, TP still maintains a certain level of response, indicating strong system resistance and environmental stability. Conversely, smaller values of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${h}_{\mathrm{max}}$$\end{document}hmax and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${h}_{\mathrm{min}}$$\end{document}hmin reflect weaker system responses under normal or extreme conditions, with the coupling relationship easily disrupted and water quality being more fragile. A larger \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta h$$\end{document}Δh value means that the amplitude of correlation fluctuations between precipitation and TP across different scales is greater, indicating that the system is more sensitive to precipitation of varying intensities and has poorer water quality stability. In contrast, a smaller \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta h$$\end{document}Δh value indicates a smoother system response, smaller overall water quality fluctuations, and higher environmental regulation capabilities. The \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${h}_{\mathrm{max}}$$\end{document}hmax, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${h}_{\mathrm{min}}$$\end{document}hmin, and Δh values calculated for the upstream and downstream of the two rivers are as follows: 2.08, 0.69, and 1.38 for the upstream of Dahong River, 1.93, 0.73, and 1.20 for the downstream of Dahong River, 2.28, 0.80, and 1.47 for the upstream of Yulin River, and 2.06, 0.61, and 1.44 for the downstream of Yulin River. The Δh values are all greater than 0 for both rivers’ upstream and downstream, indicating that the temporal evolution of the correlation between precipitation and TP exhibits multifractal characteristics. From the perspective of the impact of farmland types, areas dominated by dryland (such as the upstream of Yulin River and Dahong River) generally show larger Δh values, meaning a greater difference between \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${h}_{\mathrm{max}}$$\end{document}hmax and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${h}_{\mathrm{min}}$$\end{document}hmin. This indicates that the response relationship between TP and precipitation changes drastically, the system stability is poorer, and it is more sensitive to precipitation disturbances. This is closely related to the characteristics of bare soil, steep slopes, and low vegetation coverage in dryland. Once precipitation occurs, it quickly forms runoff and carries TP out. In contrast, areas dominated by paddy fields (such as the downstream section of the Dahong River) exhibit relatively smaller Δh values, indicating that the system’s regulation of TP is more consistent and the coupling relationship is more stable. This may be attributed to the typical water retention and slow-release functions of paddy fields, which can, to some extent, delay and modulate the phosphorus loss process. As a result, the response of TP to precipitation disturbances across different time scales becomes more uniform, leading to higher long-term stability. To verify whether the differences in the precipitation and TP relationship between the two rivers under different farmland control types are indeed significant. This study further introduces multifractal theory for investigation. Based on this theory, the multifractal nature of a time series primarily arises from two factors: long-range correlations between large and small fluctuations, and the presence of a heavy-tailed probability distribution with pronounced peaks. By randomly shuffling the original sequence to eliminate its intrinsic temporal correlations, one can effectively assess the contribution of long-term persistence to the multifractality. Additionally, by applying the phase randomization method to remove the nonlinear distribution in the original series while preserving its correlation trend, it becomes possible to evaluate the contribution of the heavy-tailed probability distribution to the multifractal characteristics. To identify the sources of multifractality in the sequences, this study performed surrogate and phase-randomization processing on the original time series. The MFDCCA method is then applied to both the randomized and surrogate sequences to obtain the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$q-h(q)$$\end{document}q-h(q) relationships (green lines for randomized sequences and red lines for surrogate sequences in Figs. 7c–d and 8c–d). The results show that the precipitation and TP series in both the upstream and downstream sections of the Yulin and Dahong Rivers exhibit the largest deviations in the randomized series. This indicates that the primary source of their multifractality arises from the long-term persistent dynamic mechanisms underlying the interactions between precipitation and TP. Temporal variability of multifractal parameters To further reveal the dynamic evolution of the correlation between precipitation and TP across different time scales. This study combines the sliding window method to analyze the temporal variability of multifractal parameters. The procedure is as follows: first, the precipitation and TP time series from January 1 to January 31, 2021 (days 1–30) are chosen as the initial window, and the first \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta h$$\end{document}Δh is computed using the MF-DCCA method. Then, the window moves forward by one day (days 2–31) to calculate the second \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta h$$\end{document}Δh, and this process continues similarly. In this manner, a total of 701 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta h$$\end{document}Δh values are obtained for each site. In this process, the addition of each new data point results in a change in the multifractal parameters. By repeatedly applying this process to the entire data sequence, with the window size gradually increasing, a set of localized measurements is obtained. Each newly calculated \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta h$$\end{document}Δh value for the precipitation and TP series is associated with the occurrence time of the last event within that window. Therefore, the variation in multifractal parameters provides information on the temporal dynamics of the precipitation and TP relationship. Figure 9 presents the 3D plots of multifractal parameters of TP and precipitation time series at the upstream and downstream sections of the Dahong and Yulin Rivers, calculated using the sliding window method. These plots show the projection of the multifractal parameters (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\lambda$$\end{document}λ, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${h}_{\mathrm{max}}$$\end{document}hmax, and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${h}_{\mathrm{min}}$$\end{document}hmin) within a 3D coordinate system. As seen in Fig. 9, there are significant differences in the multifractal parameters between the upstream and downstream sections of the Dahong River. The parameters at the upstream site are relatively scattered, while those at the downstream site are clustered together. In contrast, the multifractal parameters at the upstream and downstream sections of the Yulin River show no clear distinction. Fig. 93D multifractal parameters of TP and precipitation sequences in the upstream and downstream of the Dahong River and Yulin River are calculated using the sliding window size method 3D multifractal parameters of TP and precipitation sequences in the upstream and downstream of the Dahong River and Yulin River are calculated using the sliding window size method Furthermore, to quantitatively assess the clustering degree of the multifractal parameters in 3D space. This study employs the Quickhull algorithm (Barber et al., 1996) to calculate the convex hull volume of the 3D point set (Cornwell et al., 2006). The algorithm constructs convex hull facets recursively and forms tetrahedra by connecting each triangular facet with a reference point. The volume of each tetrahedron is calculated using the following formula:13\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$V_i=\frac16\mid\left(\overrightarrow{A_i}-\overrightarrow O\right)\cdot\left[\left(\overrightarrow{B_i}-\overrightarrow O\right)\times\left(\overrightarrow{C_i}-\overrightarrow O\right)\right]\mid$$\end{document}Vi=16∣Ai→-O→·Bi→-O→×Ci→-O→∣where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\overrightarrow{{A}_{i}}$$\end{document}Ai→, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\overrightarrow{{B}_{i}}$$\end{document}Bi→, and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\overrightarrow{{C}_{i}}$$\end{document}Ci→ are the three vertices of the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$i$$\end{document}i triangular facet, and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\overrightarrow{O}$$\end{document}O→ is the reference point (usually chosen as the origin or any point inside the convex hull). Summing the volumes \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${V}_{i}$$\end{document}Vi of all tetrahedra yields the total convex hull volume \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$V$$\end{document}V:14\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$V=\sum_{i=1}^nV_i$$\end{document}V=∑i=1nVi15where a smaller convex hull volume \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$V$$\end{document}V indicates that the point set distribution is more concentrated, meaning the multifractal characteristics of the region are more stable across different time scales. Conversely, a larger volume \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$V$$\end{document}V indicates greater variability in multifractal behavior. To compare the differences in multifractal clustering between upstream and downstream areas, this study introduces the convex hull volume difference \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta V$$\end{document}ΔV, defined as follows:where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta V>0$$\end{document}ΔV>0 indicates that the multifractal features upstream are more dispersed, while those downstream are more concentrated, that is, the upstream water quality is worse than the downstream. \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta V<0$$\end{document}ΔV<0 indicates the opposite situation, meaning upstream water quality is better than downstream. \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta V\approx 0$$\end{document}ΔV≈0 means there is little difference in multifractal features between upstream and downstream, and water quality conditions are similar. Generally, as a river flows from upstream to downstream, more pollutants are introduced along the way, resulting in better water quality upstream than downstream. If downstream water quality is better than upstream, it suggests the presence of certain underlying mechanisms in the downstream region, such as changes in farmland types or reductions in pollution sources. Therefore, the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta V$$\end{document}ΔV value not only reflects the consistency differences of multifractal structures between upstream and downstream but also serves as a quantitative indicator of how land use type differences influence the coupling behavior between TP and precipitation. The calculated convex hull volumes \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$V$$\end{document}V of the multifractal parameters for the TP and precipitation sequences in the upstream and downstream regions of the Dahong River and Yulin River are as follows: Dahong River upstream 0.71, downstream 0.02; Yulin River upstream 0.61, downstream 0.93. The difference \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta V$$\end{document}ΔV between upstream and downstream is 0.69 for the Dahong River and −0.32 for the Yulin River. It can be seen that the multifractal features in the downstream paddy field area of the Dahong River are significantly more concentrated. This suggests that the water quality in downstream of the Dahong River is superior. The relationship between TP and precipitation is more stable and regular in this region, while there is significant fluctuation in the upstream arid areas. In contrast, the aggregation difference between upstream and downstream of the Yulin River (both dryland) is smaller, even showing a more dispersed feature downstream. This suggests that the water quality downstream of the Yulin River is worse, and TP responds more sensitively to precipitation. The above results indicate that the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta V$$\end{document}ΔV calculated based on multifractal parameters can effectively quantify the impact of different farmland types on the structural complexity of the TP-precipitation relationship, holding potential value for identifying non-point source pollution and guiding land use management. Environmental significance To gain a deeper understanding of the significant differences in river TP response to precipitation under different farmland types. This study explores the impact mechanism of farmland structure on the pathway and intensity of phosphorus migration, starting from the process of “precipitation driven-farmland response-phosphorus migration.” This study finds that different types of farmland exhibit not only distinct characteristics of phosphorus loss under precipitation conditions but also significant spatiotemporal response differences, thereby exerting lasting and profound impacts on river water quality. Specifically, paddy fields are mostly located in low-lying areas and possess good water storage and regulation capacities (Shrivastava et al., 2018). During the early stages of precipitation, they can effectively retain surface runoff and adsorb part of the phosphorus, acting as a buffer (Liu et al., 2021). However, when precipitation intensifies or accumulates over time, the original water storage capacity gradually becomes saturated, and excess water begins to discharge, thereby increasing the risk of phosphorus being transported with the flow. The large influx of phosphorus into water bodies can lead to algal blooms, which in turn absorb phosphorus (Neal et al., 2006). Therefore, water quality in downstream areas controlled by rice paddies tends to be better than that upstream. In contrast, dryland areas are often distributed on steeper slopes with weaker vegetation cover, resulting in longer periods of exposed soil and rapid response to precipitation changes. They tend to generate surface runoff quickly in a short time, rapidly carrying phosphorus into rivers, making them major contributors to nonpoint source pollution. This difference between the “slow-release and lagged” response in paddy fields and the “rapid and immediate” response in drylands is a key controlling factor for phosphorus migration pathways and dynamic changes under different farmland types. Based on this, the study innovatively combines fractal methods with high-resolution remote sensing data to establish a multisource data analysis framework. It can effectively identify the coupled mechanism of “precipitation driven-farmland response-phosphorus migration” in small watersheds. Compared with traditional single-source research approaches, this framework can more comprehensively capture the multiscale regulatory effects of different farmland types and buffering water bodies on the spatial distribution and migration pathways of TP under local precipitation conditions. This significantly improves the spatiotemporal resolution and dynamic identification capability of nonpoint source pollution processes. This method not only enriches the theoretical understanding of how farmland use structure dynamically influences phosphorus migration but also provides a new technical approach for precisely identifying phosphorus migration fluxes within the complex interplay of farmland structures and hydrological processes. Practically, the results help scientifically guide the optimization of different farmland layouts, strengthen ecological regulation functions, and enhance the interception and absorption efficiency of buffering water bodies against pollutants. This offers important scientific evidence and decision-making references for the construction of coupled mechanisms between sustainable agricultural development and efficient watershed water environment management. The precipitation alone cannot fully represent hydrological processes such as flushing, dilution, and retention. Due to the absence of continuous discharge observations in both basins, river flow was not directly incorporated into the present analysis. Future research will integrate observed discharge data or hydrological simulations (e.g., SWAT-based modeling) to further strengthen the causal interpretation of rainfall-driven phosphorus migration. Conclusions This study focuses on two small watersheds in the Sichuan Basin (Yulin River and Dahong River) to systematically analyze the multiscale coupling relationship between daily precipitation and TP concentration from 2021 to 2022, as well as their response mechanisms to farmland structure. Although the two rivers exhibit highly similar meteorological conditions and pollution sources, TP concentrations show significant differences between the upstream and downstream areas. The DCCA analysis results indicate a significant power-law cross-correlation between precipitation and TP over long time scales, exhibiting long-term persistence. Specifically, the DCCA scaling exponents \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\lambda$$\end{document}λ for the upstream and downstream of the Dahong River are 0.96 and 0.98, respectively, while those for the upstream and downstream of the Yulin River are 0.76 and 1.23, respectively. Additionally, in the Dahong River downstream, which is dominated by paddy fields, the DCCA cross-correlation coefficient (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\rho }_{DCCA}$$\end{document}ρDCCA) with TP gradually increases and stabilizes around 0.8 at medium to long scales (s > 60 days). In contrast, the upstream of Dahong River and both upstream and downstream of Yulin River, dominated by dryland farming, show positive correlation with TP only within a 2-month scale (s < 60 days), followed by negative correlation with larger fluctuations. This reflects the dryland system′s rapid response and unstable output to precipitation runoff, whereas paddy fields exhibit short-term and more stable output characteristics. MFDCCA analysis further reveals the multifractal characteristics of the relationship between precipitation and TP. The \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta h$$\end{document}Δh values for the upstream and downstream of the Dahong River are 1.38 and 1.20, respectively, while those for the upstream and downstream of the Yulin River are 1.47 and 1.44, respectively. The \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta h$$\end{document}Δh values in areas dominated by paddy fields are smaller, indicating a more stable system response to precipitation disturbances and a more stable coupling relationship. In contrast, the larger \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta h$$\end{document}Δh values in dryland-controlled areas reflect stronger multiscale heterogeneity and response sensitivity. In addition, this study quantitatively evaluated the aggregation degree of multifractal parameters based on the volume of 3D convex hulls, and characterized the spatial differentiation of the TP and precipitation coupling relationship under different farmland structures through the volume difference \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta V$$\end{document}ΔV between upstream and downstream. The results show that the convex hull volumes \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$V$$\end{document}V for the upstream and downstream of the Dahong River are 0.71 and 0.02, respectively, with a volume difference \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta V$$\end{document}ΔV of 0.69. This indicates that the downstream system of the Dahong River exhibits a more concentrated coupling relationship and a more stable response. In contrast, the volumes \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$V$$\end{document}V for the upstream and downstream of the Yulin River are 0.60 and 0.93, respectively, with a volume difference \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta V$$\end{document}ΔV of –0.33, reflecting a more dispersed and unstable TP and precipitation response downstream of the Yulin River. In summary, this study develops an integrated analytical framework combining high-resolution remote sensing data, DCCA and MFDCCA analyses, and quantitative calculation of convex hull volumes to reveal the nonlinear coupling mechanism and multiscale dynamic characteristics among precipitation, farmland type, and phosphorus migration. The findings provide a scientific quantitative basis and technical support for the dynamic identification of nonpoint source phosphorus pollution in small watersheds, optimization of farmland use structure, and water environment management.