Impact of historical and future land use land cover on spatial-temporal variation of discharge and sediment load of Upper Tapi Basin, India Open Access

In the past, due to expeditious growth in human population and the change in human activities and climate, changes have been in the natural balance of basins worldwide. This change in human response towards the natural resources is attracting worldwide attention. Intensified human activity has resulted in environmental changes in larger river basins (Steffen 2004). Anthropogenic activities like alteration in land usage, construction of reservoirs and check dams, farming and agricultural practices have been related to the change in behavior of river basins (Zhao et al. 2015). Although land use changes have been significant, construction of dams and reservoirs have played a major role in the decline of discharge and suspended sediment load (SSL) of rivers (Syvitski et al. 2009). Climate holds a major contribution in controlling stream flow and sediment load through the rivers (Das & Banerjee 2021). The unbalance in the environment has led to an increase in river basin management practices. For basin management practices, sediment yield assessment and its determining factors play a major role.

Due to such variations in sediment load and discharge, abrupt changes and trends are reported by many researchers for large river basins worldwide and focus is laid on determining the causes behind it. As per Walling & Fang (2003), sediment load of the Yellow River has declined considerably due to reduced precipitation, soil and water management programs and water abstraction. Zhao et al. (2018) analyzed the sediment load for the Huangfuchuan watershed, a tributary of the Yellow River, and found a 70% reduction in sediment load due to climate change and human activities. Construction of dams resulted in clean water due to a reduction in the sediment load in the Nile and Colorado rivers (Vörösmarty et al. 2003). Panda et al. (2011) stated a sediment load reduction in the peninsular rivers of India from 1986 to 2006, while a major load reduction is observed in Narmada River, −2.07 × 10⁶ t/yr, due to the rise in number of hydraulic structures in the basin. Land use land cover (LULC) dynamics are recognized as the reason for change in sediment transport of the catchment (Kuhnle et al. 1996; Sinha & Eldho 2018).

Annual sediment load transported by Gangetic rivers is 2390 tonnes/km² while annual sediment load carried by tropical rivers is 216 tonnes/km², hence Gangetic rivers carries more load (Milliman & Meade 1983). According to studies, the development of hydraulic structures results in a decrease in sediment load. For example, Krishna River has faced a drop in sediment load and discharge due to the construction of hydraulic structures (Gamage & Smakhtin 2009). Similarly, Narmada River basin has also undergone a substantial reduction in sediment flux in which Sardar Sarovar Dam is the largest dam in the Narmada basin and it traps nearby 60–80% of the sediment load (Gupta & Chakrapani 2005).

Mann-Kendal (MK), modified Mann-Kendall (MMK) and innovative trend analysis (ITA) are different methods used to determine trends in time series. Mehta & Yadav (2021) used the MK test for trend analysis of rainfall and Sen's slope is used to determine magnitude of the trend. Pastagia & Mehta (2022) used ITA to identify the trend in rainfall of Rajsamand district of Rajasthan using graphical representation. Mehta & Yadav (2022) used the MK test to determine trends in temperature and rainfall for semi-arid and arid regions of Rajasthan. Anushka et al. (2020) used the MK test to determine trends for rainfall and temperature for the island of Trinidad and Tobago. These studies do not state the variation in discharge and suspended sediment load. Adib et al. (2021) used the MK test and Sen's slope to predict annual and monthly discharges and used Markov chain to generate time series data. Adib & Mahmoodi (2017) used an Artificial Neural Network (ANN) Genetic Algorithm (GA) model to predict SSL along the Markov chain. Sharma et al. (2018) carried out trend analysis of rainfall and temperature for Upper Tapi Basin using MK and MMK tests. The increase in frequency and magnitude of extreme rainfall in the basin has been attributed to the increasing trend in maximum and minimum temperatures, reducing forest cover and rapid pace of urbanization. Sharma et al. (2019) analysed the impact of rainfall variability and anthropogenic activities on streamflow variation of Upper Tapi Basin.

LULC dynamics show the historical land changes over the period of time, while LULC change modelling deals with the interpolation or extrapolation from the given LULC information for a given time period. Change detection through LULC prediction is helpful in identifying urbanisation, landscape changes etc. In order to inculcate proper resources management and decision making, it is necessary to understand changing landscape pattern, influence of human activities, and utilization of resources. Several works have been conducted using a land change modeler (LCM) modeler to predict LULC to identify the impact of human caused activities on the natural habitat (Liping et al. 2018). Changes occurring in the environment are a result of manmade interventions. Gupta & Sharma (2020) and Anand & Oinam (2020) carried out LULC analysis and its prediction was carried out in Terrset using LCM and the results indicated the impact of anthropogenic activities.

Previous research did not take into account the trend analysis and variations in flow and sediment load associated with relative LULC in the Upper Tapi basin over the decades. Hence, the aim of the present study is to detect the trend in sediment load and discharge in the basin along with the impact of decadal variation of LULC. Trend detection was carried out using MK, MMK test and ITA. LULC changes were investigated using GIS and future projection of LULC for 2030 was carried out using IDRISI Terrset. The study's conclusions will be valuable for the long-term management of natural resources.

The study focuses on sediment load variation in Indian rivers, and analysis of sediment load variation was carried out for Tapi River which is a major west flowing river of India. It is a west flowing river considered to be a major river of Peninsular India and it is 724 km long. The Tapi is the second largest interstate river basin that drains westward originating from Multai, Betul district, Madhya Pradesh.

The Tapi Basin is located in the Deccan Plateau's northern region and covers 65,145 km², or roughly 2% of the country's entire geographical area. Maharashtra accounts for over 80% of the basin's area. The Satpura range runs through the north, the Mahadeo hills go through the east, the Ajanta range and Satmala hills run through the south, and the Arabian Sea runs through the west. The Tapi River and its branches flow across Vidharbha, Khandesh, and Gujarat plains, as well as significant swaths of Maharashtra and a minor portion of Madhya Pradesh and Gujarat.

In the present study, the discharge and sediment load data were collected from the Central Water Commission (CWC) and Landsat imageries were downloaded from United States Geological Survey (USGS) and their description is mentioned below.

The amount of sediment load transported and soil erosion, along with the discharge, primarily depends on how the land is used presently in the basin (Valentin et al. 2008). In the present study, satellite images from the years 1989, 1999, 2009 and 2020 were retrieved from USGS Earth Explorer. Landsat 5 and Landsat 8 images were downloaded and LULC maps were prepared using supervised classification to understand the decadal variation. Table 2 provides a summary of the data used for image classification.

The widely acknowledged MK test, MMK test, Sen's slope, and ITA were used to perform spatial and temporal variation of discharge and SSL. Figure 2 depicts the methodology used in this study.

Here, d indicates the slope i.e., magnitude and X_j and X_i are values at times i and j, and n is the number of variables.

Pettitt (1979) is one of the convenient approaches to find abrupt change in a time series data. Suppose, X₁, X₂, …. X_t, is a set of random variables having change point at location τ (X_t for t = 1, 2, …, τ follows a mutual distribution function F₁(x) and X_t for t = τ + 1,., T have a common distribution function F₁(x) where both functions are different from each other. There will be no change point in the time series if the variable follows the same distribution (null hypothesis), while if change point occurs, a different distribution exists in the variable (alternative hypothesis).

Here, is the trend slope for the given time series. A positive value of Sen's slope shows an increasing trend while negative values show the declining trend. It is worth noting that calculating the MMK test necessitates estimating the Sen's slope.

Here, the value of D indicates the trend, n is the whole number of data, x_i and y_i are first and second sub series respectively. If n is an odd number, then the first value is discarded to obtain an equal amount of data in the first and second sub series. Alashan (2018) gave change boxes methodology for trend representation of scattered points in ITA. For the scattered points, the data are divided into three parts; low, medium and high value groups arranged as per experts review or by dividing the time series strictly by considering first half series as X, the mean X_M and S_x as the standard deviation. Using these three terms, ranges are defined for low, medium and upper ranges as X < X_M – S_X; X_M – S_X < X < X_M + S_X and X > X_M + S_X, respectively.

Maximum Likelihood Classification (MLC) was used to classify the datasets and five classes such as water bodies, built-up area, forest, agricultural land and barren land were distinguished by giving training samples. Google earth visualization was used to avoid misidentification of training samples of LULC classes.

Accuracy assessment was carried out in order to check how accurate the classified image is with respect to the ground truth. Failure to attain the expected accuracy is generally construed as a deficiency of satellite data classification contrary to LULC (Abijith & Saravanan 2021). In order to determine the accuracy of LULC, a confusion matrix was generated by selecting samples from each class of 1989, 1999, 2009 and 2020.

In the present study, Land Change Modeler (LCM), available in IDRISI Terrset software, was used for LULC modelling and prediction. LCM in Terrset was used to determine spatial changes in the area, and to predict and validate the LULC. The LCM modeler uses different methods like Multi-Layer Perceptron (MLP), Support Vector Machine (SVM), and Random Forests (RF) to model changes in LUCL. To run the LCM model, input for two different time period i.e., 1999 and 2009 in present study, must have some format, the same spatial extent, same resolution, the background should be same in both maps and have a value of zero, legends of both maps should be same. (Abuelaish & Olmedo 2016). The LCM modeler is divided into three parts listed as change analysis tab, transition potential tab and a prediction tab. The change analysis tab gives details of changes occurring in each category of classification over the period. The transition potential tab models each sub transition model or group of sub model depending on the inputs needed by the models. To predict the LULC, the selected sub models should be modelled by choosing the appropriate method. In the present study, a combination of SVM with Markov chain analysis was used. Before modelling the sub transition model, it is necessary to identify the drivers which influence the chosen classes (Sharjee et al. 2016).

Markov Chain Model is based on the sequence of events to occur which will be based on the occurrence of previous events. In LULC changes, matrices are used by Markov chain to represent the transition occurring in the land use categories. Markov chain uses the transition from one period to another period to predict the future LULC.

Highest monthly mean discharge is seen in the month of August while maximum monthly discharge is observed in the month of July in Burhanpur G.S, while maximum monthly mean sediment load and maximum monthly is observed in July. It can be seen that Gopalkheda G.S. carries less sediment load and discharge as compared to Yerli G.S.

For Yerli G.S., maximum discharge and maximum mean monthly discharge is obtained in August and the same result is obtained for sediment load. For Gopalkheda G.S., maximum discharge and maximum mean monthly discharge is obtained in August and the same result is obtained for sediment load. Yerli G.S. tends to carry more sediment load and discharge than Gopalkheda G.S.

Results of trend analysis obtained from MK and MMK tests for discharge in UTB are shown in Table 3. Burhanpur G.S., Gopalkheda G.S. and Yerli G.S. have shown a decreasing trend in daily discharge using the MK and MMK method. All the stations are showing a significant decreasing trend (z < −1.96, for 5% confidence interval). It is seen from Table 2 that the MK test detected more significance as compared to that of the MMK test because autocorrelation is not removed in the data while doing the MK test. Yerli G.S. showed the most significant decreasing trend in discharge in Upper Tapi Basin. Tapi and its tributary Purna have rainfall in terms of water source. Thus, a decrease in discharge may occur due to a decrease in rainfall. Sharma et al. (2018) analyzed the rainfall trend over UTB and found a decreasing trend over the basin for total annual rainfall. Burhanpur, Gopalkheda and Yerli have shown a decreasing trend for discharge; Yerli (z = −7.68) has shown a greater decreasing trend compared to the other two stations.

Burhanpur G.S. and Yerli G.S. have shown a decreasing trend in daily SSL using MK and MMK methods. All the stations are showing a significant decreasing trend (z < −1.96, for 5% confidence interval). It is seen from Table 5 that the MK test detected more significance as compared to that of the MMK test because autocorrelation was not removed in the data while doing the MK test. Yerli G.S. showed the most significant decreasing trend in SSL in Upper Tapi Basin. Gopalkheda G.S. showed an increase in sediment load.

Figure 7 shows the change point analysis using the Pettitt test for Annual Water Discharge (m³) (AWD) at Burhanpur G.S. AWD is the total volume of discharge passing throughout the year. The change point of discharge at Burhanpur G.S. is observed in 1988. For Gopalkheda G.S., the change point is observed in 2017 and for Yerli G.S. it is observed in the year 2000. Figure 7 also shows the change point analysis using Pettitt test for Annual Sediment Load (M.T.) (ASL) at Burhanpur G.S. ASL is the total suspended sediment load carried throughout the year. At Gopalkheda G.S., the change point for ASL is obtained in 2016, for Yerli G.S. in 1990 and at Burhanpur G.S. in 1999.

Figure 8 shows the double mass plots which are developed for the G.S. to understand the change in ratio of sediment load and discharge. Gopalkheda shows steeper regression before the change point, i.e., 3.16, while it declines to 1.57 after occurrence of the change point. For Yerli G.S., steeper regression before the change point is observed, i.e., 4.9, while it declines to 2.57 after occurrence of the change point. Slope represents the ratio of sediment load and discharge going downstream on Purna River, the slope decreases from Yerli to Gopalkheda indicating a decrease in sediment load. For Burhanpur G.S., the slope decreases from 1.57 to 1.40. Burhanpur G.S. has observed less change in regression slope as compared to G.S. at Purna River, so lesser change in sediment load has occurred. Comparing all the G.S. at UTB, Yerli G.S. holds maximum regression slope after the change point, thus it carries more sediment load throughout the basin.

Table 7 gives the area statistics of historical variation of LULC. From 1989 to 2020, the vegetation in the basin decreased from 43 to 21% of the basin area, while agricultural land increased from 50 to 60.79%, and the built-up area increased from 1 to 8.3% of the total area. Water bodies of the basin increased to 2.1% from 1.2%. Reduction in the vegetation is observed which has converted into agricultural land because of the increase in the urban area, giving rise to demands of population.

It is observed from the trend analysis that sediment load has shown a decreasing trend at Yerli and Burhanpur G.S. while an increasing trend is observed at Gopalkheda G.S. and discharge has shown a decreasing trend at all G.S. LULC has also shown drastic changes from 1989 to 2020. Thus, it can be said that the change in LULC of the basin has impacted the discharge and sediment load pattern of the basin. Hence, in order to correlate the LULC changes, sediment load and discharge patterns were studied to determine the trend. A decrease in vegetation and increase in agricultural land has increased the sediment load in Purna River at Gopalkheda G.S. These changes have affected the sediment load carried by the river, soil erosion, and streamflow in the basin. A reduction in vegetation and simultaneous growth in agricultural land increases the sediment yield of the basin (Sinha & Eldho 2018).

An accuracy assessment of classification was carried out to compare the classified data to the ground truth data. Table 8 shows the accuracy assessment of LULC.

To validate the projected LULC of 2020 with the observed 2020 LULC, Kappa coefficient is used. To calculate the Kappa coefficient, first the LULC is predicted and then by forming a confusion matrix accuracy is assessed. Overall accuracy of 82.35% was obtained with a Kappa coefficient of 78.12%.

The Cellular Automata-Markov Chain works on the principle of past trends to predict future scenarios. Land Change Modeler in Terrset implements the CA-Markov method and later the transition potentials are run and optimized by Support Vector Machine (SVM). After giving LULC of 1999 and 2009, a transition sub model was obtained from the change maps. After obtaining the LULC 2020 from 1999 to 2009 using SVM, same methodology will be used to predict LULC of 2030. Using images of 2009 and 2020, LULC of 2030 was predicted. Figure 10(a) and 10(b) show the predicted 202 and 2030 image of Upper Tapi Basin respectively. The obtained LULC 2030 is of 30 m resolution.

Due to rapid urbanization, the increase in built-up area leads to an increase in impervious surfaces. Because of this, the rate of infiltration decreases, hence an increase in frequency of high streamflow is observed and low streamflow decreases (Wilson & Weng 2011). Due to increasing anthropogenic activities, the built-up area is increasing with a decrease in agricultural land and vegetation in the basin. For Tapi River at Burhanpur G.S., a decrease in sediment load will be observed due to the decrease in vegetation and agricultural land. For Purna River, as per the decadal variation, the increase in agricultural land has increased the sediment load but for projected LULC of 2030, agricultural land is decreasing from 17,654 to 16,843 km², which may lead to a decrease in sediment load.

The present study focuses on trend detection of daily discharge and sediment load in Upper Tapi Basin using Mann-Kendall, Modified Mann-Kendall, and Innovative Trend Analysis. Change point for Annual Sediment Load and Annual Water Discharge is obtained through Pettitt Test. Decadal analysis of LULC is carried out from 1989 to 2019. The changes in LULC pattern are related with discharge and sediment load pattern. The findings of the study are as follows:

• The spatial-temporal variation in discharge in the UTB is mainly controlled by climatic variation, i.e, due to variation in precipitation, as major discharge is observed in the months of monsoon. Thus, rainfall is the only source of water in the basin.

• The variation of sediment load over the basin turns out to be a complex function depending upon discharge, rainfall, anthropogenic activities, and changes in LULC. Although discharge is the carrier of sediment load, due to change in geological conditions, sediment load may get deposited.

• A significant decreasing trend for discharge is observed using MK and MMK tests at all three G.S. ITA gives clear trend representation of the time series using change boxes. On average, discharge in UTB has shown a decreasing trend for middle ranges while upper values of discharge have shown an increasing trend. Maximum discharge in UTB was observed during the flood in 2006. A similar trend was observed for sediment load.

• The slopes of sediment load and discharge in double mass plot aligns with the results obtained from the trend analysis. The decline in slope at all three G.S. implies that a decrease in sediment load is more significant than discharge in the basin. Major change in slope is observed at Yerli G.S. as compared to Gopalkheda and Burhanpur G.S., thus a significant reduction in sediment load is observed at Yerli G.S.

• From decadal analysis of LULC from 1989 to 2020, vegetation in UTB has decreased and an increase in agricultural land and built-up area is observed. This change in LULC is due to urbanization, increase in population and their requirements. These changes have led to a reduction in discharge and sediment load of the basin.

• Using LCM Modeler in IDRIDI Terrset, LULC of 2020 was projected and validated with overall accuracy and Kappa coefficient of 82.35 and 78.12 respectively. Future LULC for Upper Tapi Basin for 2030 is projected and water-bodies, built-up area and barren land have increased in contrast to the decrease in vegetation and agricultural land. The rise in population, increase in the built-up areas and decrease in the vegetation is a threat to a sustainable environment. The impact of anthropogenic activities on land use changes can be managed by afforestation, water conservation measures and sediment management in the basin area.

Data cannot be made publicly available; readers should contact the corresponding author for details.

The authors declare there is no conflict.