The aim is to use observations and simulations by a spatially distributed model for analysing the role of spatial organization of terraces, ditches and arrangement of various tillage practices on the complex behaviour of a small Mediterranean basin.
Abstract:
Spatial Organization and Complex Behavior of a Small Agricultural Mediterranean BasinRoger Moussa1, Francois Colin 2
1 INRA, UMR LISAH, 2 Place Pierre Viala, 34060 Montpellier Cedex 1, France, moussa@supagro.inra.fr
2 Montpellier SupAgro, UMR LISAH, 2 Place Pierre Viala, 34060 Montpellier Cedex 1, France, colinf@supagro.inra.fr
1. Introduction
Runoff processes have been largely modelled in natural catchments. However, on agricultural catchments, the intensity of processes and so runoff magnitude differ since the agricultural land use, the division of the landscape in fields, terraces and the ditch network, are significant factors in controlling flood generation.
The agricultural field can be considered as the smallest spatial unit in agricultural landscapes, since it is homogeneous with respect to crop type and agricultural operations applied to it, such as soil tillage. Agricultural operations like tillage have influence greatly the local surface runoff, infiltration and surface storage by altering soil hydrologic properties and soil surface roughness. Tillage may increases infiltration by increasing soil porosity, but in some instances, tillage can also reduce infiltration since it may disrupt the continuity of pores between the top and subsoil.
Moreover, in large parts of the Mediterranean region, hillslopes in agricultural basins are characterised by terrace cultivation, an agricultural practice by which parts of the hillslope are levelled in a direction perpendicular the overall gradient of the terrain. The resulting agricultural terraces increase arable land surface, and it is generally assumed that they also reduce the risk of erosion.
Also, the ditch networks influence the water transfer from the fields to the catchment outlet. In comparison to natural drainage networks, they modify water transfer from fields to the catchment outlet in two ways. First, since the ditches follow the field limits, water flow doesnot necessarily follow the steepest slope of the catchment surface topography. Thus, we can expect that the ditch networks modify the average distance and slope between the fields and the catchment outlet. Second, since the ditches are excavations in the soil, they influence largely the flow exchange processes between the surface and the groundwater.
In the past, the spatial organization of fields, the introduction of terraces and ditch networks, and the choice of soil surface treatment techniques were mainly decided to facilitate the agricultural operations and increase the profitability of farming systems. Now, with the increase of environmental concerns about the occurrence of severe flood events, erosion processes, and the pollution of surface waters by pesticides, the question arises as to how much the spatial organisation of an agricultural catchment influences flow of water and matter at the catchment scale?
In this paper, we address this later question by a simulation approach that uses a spatially distributed hydrological model that attempts to take into account the main characteristics of agricultural catchments. The objective is to analyse the role of spatial organization on the complex behaviour of a small agricultural Mediterranean catchment: i) the role of terraces; ii) the role of the ditch network, and iii) the impact of the spatial arrangement of various tillage practices;
In the following, we present the main characteristics of the studied catchment and the hydrological model used and describes the results of scenarios testing the role of terraces, the role of the ditch network and the impact of various tillage practices.
2. The study site and the model
2.1.The study site
The Roujan basin is an experimental catchment (0.91 km2) located in Southern France. Annual rainfall varies between 500 and 1400 mm. Rainfall is usually of high intensity and short duration.
The catchment is mainly covered by vineyards (0.71 km2) and is divided into 237 fields. A survey identified two main soil treatments for weeding. In one, herbicides are applied over the whole field without any tillage. In the other, the soil is tilled one to three times during the growing period between March and July. On the Roujan catchment, the soil surface features of vine field parts were classified in two types, non-tilled field and tilled field. The drainage network is formed by man-made ditches and generally follows agricultural field limits. Their density is variable. Typically, they are 0.7 to 1.2 m wide and 0.8 to 1.4 m deep. The total length of the ditch network is 11069 m. Terraces in this catchment have been associated with terrain slope, pre-existing terraces and traces of ancient human habitation. A field study demonstrated that around 60% of the fields are surrounded by terraces, generally walls of 1 to 3 m height and 1m width.
The basic instrumental design, set up in May 1992, consists of a meteorological station, rain gauges, stream flow recorders, piezometers and tensio-neutronic sites. Four tipping-bucket rain gauges are used to monitor rainfall with a time step of one minute. In an attempt to describe spatial variability of runoff, discharge is measured at four gauging stations using a Venturi channel: two at the outlet of two fields (one non-tilled 1200 m2, and one tilled 3240 m2), one at the hillslope scale grouping six fields (5200 m2), and the main station at the outlet (0.91 km2). The time step of measurements is one minute. A network of piezometers (3 or 5 m deep) has also been installed to measure the spatial variation of the water table level on weekly basis.
The major runoff events are usually caused by high-intensity short-duration storms. The outflow at the outlet is event-dominated and runoff coefficient at the catchment scale varied from 0 to 68% with a lag time of 30 to 60 min. The catchment''s very short response time to rainfall pulses suggest that runoff during rainfall is primarily dominated by Hortonian overland flow. The predominance of overland flow is very closely linked to the climatic and soil characteristics (capping soils) but also to farming methods (tillage practices producing relatively impervious soil surfaces). This produces intense runoff processes at the field scale (e.g. coefficients measured at the outlet of the non-tilled field can reach 70%).
2.2. The model
This study uses MHYDAS (acronym for Modelisation Hydrologique Distribuee des Agrosystemes, which is French for Distributed Hydrological Model for Agricultural; Moussa et al., 2002), a distributed model developed for simulating storm flow in agricultural catchments, which are characterized by a complex spatial organization. The spatial organization is represented in the model by a subdivision into hydrologic units and channel sections in the form of objects, for each of which state variables are calculated. Hydrologic units have a geographic extent defined by field limits and drains. The connection between different spatial units resembles a tree-like structure that reflects the main drainage pattern and the catchment topography. MHYDAS calculates for each unit the infiltration rate and overland flow. Evapotranspiration is not accounted for, because the purpose of the model is to simulate storm events. The model structure used in MHYDAS allows for hydrologic simulations of complex agricultural catchments with spatial units of virtually any shape and with the possibility to account for temporal variations in hydrologic properties that result from agricultural practice.
MHYDAS uses an infiltration model (e.g. Green and Ampt, Parlange, Richards) to partition rainfall into infiltration and overland flow. The resulting infiltration excess flows off along the land surface as overland flow. Parameters used by this model are the hydraulic conductivity at natural saturation, the soil water content at natural saturation, and the effective capillary drive, which depends on initial water content. On each hydrologic unit, overland flow is routed using the diffusive wave equation, which has two parameters, the flow celerity and the flow diffusion. A similar approach is used to route the hydrograph through the channel network to the outlet. The discharge calculated with the diffusion wave approximation is converted into water height y using Manning's equation. Finally, channel infiltration is accounted for using Darcy's equation, function of the difference of water levels in the channel network and in the groundwater table. A multi-criteria calibration strategy for MHYDAS was developed and applied at various scales, the plot, the hillslope and the whole basin (Hallema et al., 2013).
3. The role of terraces
In order to study the effect of the spatial configuration of the hillslope on hydrologic response MHYDAS was applied on the experimental hillslope (5200 m2; 6 fields) (Hallema and Moussa, in press) on eight flood events: four events for calibration and four for validation. Then, three hypothetical hillslope configurations were defined in order to evaluate the effect of terrace cultivation and channel network configuration on the hydrologic response of the experimental hillslope:
Scenario A, hillslope with natural drainage: in this scenario, there are no terraces or drainage channels. The channel network is extracted from a DEM instead and resembles a natural stream network with channels that follow the direction of the steepest slope.
Scenario B, hillslope with increased drainage density: this scenario is based on the field situation with one additional transverse channel per each field), resulting in a total of 12 fields.
Scenario C, hillslope with increased drainage density and more terraces: this scenario is like scenario B, but with the difference that each of the 12 original and newly created fields is located on a different terrace. In the model, this is represented by a reduced hillslope gradient (now 2%, was 12% in the original situation).
We simulated the three scenarios with the calibrated parameter values obtained previously. A comparison with the hydrograph simulated for the original situation leads to the following findings: The peak flow simulated for the hillslope with natural drainage (scenario A) was 11 to 23% higher during all of the eight rainfall-runoff events, and occurred 3 to 5 min earlier compared to the original hillslope with six terraces and artificial drainage. The lag time simulated for the original hillslope was 15 min.
For scenario B (hillslope with increased drainage density), the travel time of overland flow was shorter while the travel time in channels increased due to the higher drainage density. We simulated a peak flow that was not quite as high as the peak flow for scenario A, but with a 4 to 12% increase still higher than the peak flow simulated for the original hillslope. The time of peak was advanced by 1 to 3 min, which indicates that increasing the drainage density leads to a faster hillslope response.
Scenario C demonstrates the effect of increased drainage density in combination with more terraces: the travel time of overland flow increased compared to the other scenarios. Peak flow was 13 to 23% lower than on the original hillslope and was delayed by 5 to 8 min. This is explained by the lower overland flow velocity (0.05 m. s1) resulting from the decrease in hillslope gradient (2% instead of 12%).
4. The role of the ditch network
MHYDAS was then used to study the influence of the ditch network on catchment runoff during typical hydrological situations with respect to tillage practices and water table depth. For this, three extensively studied flood events were chosen corresponding to: i) one situation in spring just after tillage with a high water table (event of 30 April 1993), ii) one intermediate situation in summer, when the incidence of tillage practices on runoff has decreased and the water level is low (event of 31 August 1994), and iii) one situation in autumn when most soils are crusted and the water table is high (event of 30 September 1994).
The actual man-made ditch network and field structure were considered as reference. The tested hypothetical scenario D supposes that no observations were made on the catchment and that the channel network was extracted automatically from DEMs. In this case, the channel network follows the steepest slope. In these applications, the DEM was established from aerial photographs with a 2 m resolution and all the channel network extremities drain the same source area (equal to 1200 m2 as this value corresponds to the mean area of all source fields in the actual reference system). The differences between the reference scenario and scenario D are the geometrical properties of the reaches, and the spatial distribution of the hydrological units. In the reference scenario, hydrological units correspond to field parts, while in scenario D they correspond to subcatchments. Consequently, the spatial configuration depicted is completely different between these two scenarios, although the respective proportions of all land use types are the same.
For these simulations, we used the parameters calibrated before. Results show that for the three typical flood events, the form of the calculated hydrograph at the catchment outlet is clearly different between the two scenarios. The time position of peak discharge is delayed of about 10 to 20 minutes (the basin response time can vary between 30 and 60 minutes). The peak discharge is reduced for the three flood events, 30/04/1993, 31/08/1994 and 30/09/1994, by - 9 %, - 43 % and -10 %, respectively. This is essentially due to the shape of the channel network, which is more sinuous in scenario B while the man-made ditches network has linear reaches. Consequently distances are longer and slopes are lower in scenario D. In this case flow celerity decreases and the lag time increases.
5. The role of tillage practices
Last, our concern was the potential impact of spatial arrangement of the two agricultural weeding practices on the catchment runoff: tilled and non-tilled (Chahinian et al., 2006a, 2006b). Particularly, we investigated what would be the extremal impacts that one could wait for. We considered that the hydrological impacts could be simulated by MHYDAS. However, the locking methodological point was how to define the extremal scenarios, knowing that the number of possible combinations is prohibitive.
The proposed method is based on a water particle tracking routine that has to be plugged-in a distributed rainfall-runoff model (Colin et al., 2012). This routine allowed to rank fields and as a consequence to build extremal scenarios of spatial arrangements. The method was then used to define the magnitude of the impact of different spatial arrangements scenarios for various catchment area ratio of the weeding practices. Finally, we proposed to use this method in a manager perspective in order to limit the peak flow at the catchment outlet for given climatic conditions. The routine allows for the calculation of two dimensionless indices for each field as follows: i) M, referring to the contribution in volume to the outlet. M values range from 0, if all the surface runoff from a field is infiltrated before joining the outlet, to 1, if all surface runoff contributes to the outlet discharge. ii) T, referring to the travel time of the surface runoff between the field and the outlet. T values range from 0, if the surface runoff of the field immediately joins the outlet, to 1, for the latest field contribution (time of concentration).
Running the routine n times allows for the calculation of the M and T indices for each of the n fields and, therefore, allows them to be ranked according to their contribution to the outlet discharge in terms of both volume and time. Assigning the highest local runoff conditions to the fields first ranked (respectively ranked last) according to M leads to a scenario that tends to maximize (respectively minimize) outlet stream flow. Assigning the highest local runoff conditions to the fields first ranked (respectively ranked last) according to T leads to a scenario in which the peak rate occurs rapidly (respectively slowly). Therefore, these two indices were useful to build extremal scenarios for a given area ratio of a land use/practice.
Hereafter, only two types of land management practices, which induced local maximum and minimum runoff conditions, were considered: tilled and non-tilled. Simulations were conducted on a median event that occurred at the end of spring/beginning of summer, at a time when infiltration in fields was spatially heterogeneous because of the weeding practices performed by farmers before the event (the event of 6 June 1997) (Colin et al., 2012). For both practices the catchment area ratios ranged between 0 and 1. For a given weeding practice area ratio, the spatial arrangements were constructed to reach the following two different goals representing the worst cases with respect to environmental issues: (1) to maximize the catchment runoff, and (2) to minimize the flood arrival time at the catchment outlet.
The simulated catchment outlet discharges are close to 0 if tilled practice catchment area ratio ranges between 0.84 and 1. Considering a non-tilled practice weeding area ratio of 0.5, the runoff coefficient and peak runoff show a maximum variation between 0.03 and 0.08 and between 0.12 and 0.39 m3 s-1, respectively. Compared to the total range obtained with a variation between 0 and 1 in the non-tilled area ratio, these changes represent more than one third of the runoff coefficient range (0-0.14) and less than half of the peak discharge range (0 to 0.6 m3 s-1). Results reveal that the non-tilled area ratio constitutes the main factor in explaining the water volume at the catchment outlet. The spatial arrangement of these practices is a significant factor in deriving a median area ratio. The conclusions for the catchment lag time are different; the average lag time is not influenced by the practice area ratio. Spatial distributions, however, can significantly modify the lag time, with the trend increasing for simulated low-flow events that correspond to a low non tilled practice area ratio.
6. Conclusion
Surface runoff on cultivated catchments varies in time and space due to the interaction between agricultural land use (e.g. fields, terraces, ditches, agricultural practices) and climatic conditions. During flood events, these elements constitute hydrological discontinuities that influence runoff contributing areas and pathways for surface runoff. The aim of this paper was to study the role of the spatial organization of these hydrological discontinuities on the complex behaviour of an agricultural Mediterranean basin. For that, the spatially distributed hydrological model MHYDAS, which is adapted to take into account the spatial organization of agricultural catchments, was parameterized on the Roujan catchment and was applied to analyse the impact of different land use scenarios.
The simulations on the Roujan catchment showed the sensitivity of the model and the impact of terraces, the ditch network and tillage practices spatial configuration on hydrological processes and runoff magnitude.
At the field scale, soil treatment has an important incidence on simulated runoff volumes at both field and catchment scale. Calibration on field runoff data indicated that tillage decreases runoff coefficients and increases infiltration. The most sensitive parameter of MHYDAS is the saturated hydraulic conductivity, which varies in space and in time according to tillage practices. These results point out that it is essential to take into account the influence of soil treatment when simulating hydrological processes at the scale of agricultural catchments.
At the hillslope scale, we were also able to analyse the effects of the different spatial configurations of a real-world agricultural hillslope and demonstrate that terrace cultivation delays the response time and reduces peak flow with respect to an undisturbed hillslope. An important consequence of this outcome is that the risks of erosion and crop damage are also significantly reduced.
Moving to the catchment scale, the ditch network constitutes man-made linear elements in the landscape, and appears to serve various purposes with regard to water flow. Its role was studied by comparing the actual man-made ditch network to a hypothetical channel network automatically extracted from a DEM and following the steepest slope. The simulations show two main results. First, the ditch network accelerates runoff by concentrating flow and avoiding natural obstacles. The hydrograph simulated using the channel network automatically extracted from the DEM doesnot allow a good representation of the value and the time of occurrence of peak discharge. The second result deals with the role of ditch networks in the flow exchange between surface and groundwater. The exchange coefficients between the ditches and the groundwater depend on infiltration properties of the ditch network, which are generally unknown, and should be measured at different locations of the ditch network. The Manning roughness coefficient can vary in space and time due to the evolution of the vegetation cover in the ditch network.
Finally, the area ratio of the two practices, tilled and non-tilled, is the main factor influencing the catchment runoff. The spatial arrangement allows modulating the catchment runoff coefficient for median area ratio. The methodology proposed herein is useful for simulating both distributed hydrological models sensitivity analysis of parameters spatial arrangements, and comparing different land use configurations in order to propose country-planning schemes.
7. References
Chahinian N., Voltz M., Moussa R., Trotoux G., 2006a. Assessing the impact of hydraulic properties of a crusted soil on overland flow modelling at the field scale. Hydrological Processes 20: 1701-1722.
Chahinian N., Moussa R., Andrieux P., Voltz M., 2006b. Accounting for temporal variation in soil hydrological properties when simulating surface runoff on tilled plots. Journal of Hydrology, 326(1-4): 135-152.
Colin F., Moussa R., Louchart X., 2012. Impact of the spatial arrangement of land management practices on surface runoff for small catchments. Hydrological Processes 26: 255-271.
Hallema D.W., Moussa R., Andrieux P., Voltz M., 2013. Parameterization and multi-criteria calibration of a distributed storm flow model applied to a Mediterranean agricultural catchment. Hydrological Processes 27: 1379-1398.
Hallema D.W., Moussa R. A model for distributed GIUH-based flow routing on natural and anthropogenic hillslopes. Hydrological Processes, 22p, In Press.
Moussa R., Voltz M., Andrieux P., 2002. Effects of the spatial organization of agricultural management on the hydrological behaviour of an agricultural catchment during flood events. Hydrological Processes 16: 393-412