Au cours des trois dernieres decennies, l’agronomie a ouvert de nombreux fronts de recherche et d’innovation lies a la prise en compte d’entites spatiales et organisationnelles de dimension superieure a celles de la parcelle et de l’exploitation agricoles. Ces progres semblent pouvoir etre valorises et stimules dans le cadre de projets d’amenagement et/ou de developpement tres divers, mis en œuvre par des collectifs d’acteurs locaux. Le territoire est aujourd’hui couramment invoque par les agronomes comme « echelle » a laquelle doivent desormais s’appliquer un grand nombre de leurs demarches. Doit-on pour autant tenir pour acquis que l’agronomie contribue au developpement territorial ? Dans l’article qui suit, on cherche a repondre a cette question, en situant l’etat de l’art en agronomie par rapport a un gradient de progression vers la prise en compte des notions de territoire et developpement territorial, au sens que donnent a ces termes les sciences humaines et sociales, et en particulier la geographie. On schematise ce gradient comme la succession de trois etapes, correspondant a des sauts qualitatifs quant au degre d’elaboration des methodes et outils a mettre en œuvre : du site a l’espace, de l’espace a la gestion de l’espace, de la gestion de l’espace au developpement territorial. Ce parcours amene au diagnostic que l’agronomie a d’ores et deja investi la gestion de l’espace, mais n’a pas encore franchi la troisieme etape, qui lui permettrait d’etre reconnue et sollicitee pour sa contribution potentielle au developpement territorial. En partant de l’analyse de trois types de documents de projets, on tente alors d’identifier quelques pistes pour concretiser cette perspective, a benefice reciproque pour l’agronomie et le developpement des territoires.

Runoff and erosion cause frequent damage through muddy floods in the loess belt of Northern Europe. One possibility for reducing damage is to lower runoff on agricultural land by spatially alternating different crops at the catchment level. But crop location results from decisions taken at the farm level. This study aimed to assess the existing leeway to modify crop location in the farms of a catchment, in order to reduce runoff at the catchment's outlet. The case study was the Bourville catchment (1086 ha), cultivated by 28 farmers and located in Pays de Caux, France. First, crop location rules in the 14 main farms of the catchment were analysed on the basis of surveys carried out with farmers, distinguishing spatial constraints from temporal ones. These rules made it possible to simulate crop location on each farm territory for the 2001-2002 crop year. Each field of the catchment was classified depending on whether one or several crops could be sown, taking into account both field history and farmer decision rules. Then two extreme scenarios of crop location in the Bourville catchment were built. Runoff simulation at the outlet with the STREAM model showed that runoff could be reduced while sticking to current farmer decision rules in terms of crop location. Depending on rainfall event characteristics, runoff reduction varied between 13.5 per cent and 4.5 per cent. Copyright (c) 2006 John Wiley & Sons, Ltd.

In the loam belt of Northern Europe and in the North-western Paris Basin, runoff and erosion may be reduced by changing agricultural practices in the agricultural catchment where runoff is produced. However, such changes may be limited by work planning constraints at the farm level. This study therefore aimed to assess farmers’ leeway to introduce stubble ploughing and mustard (Brassica alba) sowing during the intercrop period. The case study was the Bourville catchment (1086ha), cultivated by 28 farmers and located in Pays de Caux, France. Firstly, workload was analysed and sequences of cultivation operations were simulated for an average climate for the 14 main farms located in the catchment. This analysis was based on surveys carried out with farmers. It was then possible to determine the number of working days available to perform stubble ploughing and mustard sowing, both of them being efficient practices to reduce runoff at the field level. Results showed that farmers did not have the same leeway as some of them were even unable to perform stubble ploughing or mustard sowing. Then at the catchment level, the impact of changes in agricultural practices on runoff volume at the catchment's outlet was simulated by using the STREAM model. These simulations showed that not taking into account work planning constraints within farms led to overestimating both the possibilities of implementing changes in farms and the efficiency of such changes on runoff reduction.

This paper explores, from an agronomic viewpoint, the way soil surface characteristics may be managed for soil and water conservation purposes in a silty loam soil context (Pays de Caux, France). We first consider the tools used to characterize the impact of farming practices on soil surface characteristics and, as a result, on runoff and erosion processes. The way farming practices modify soil surface characteristics must be considered at various spatial scales, from the metric pattern produced by farm machinery, to the network of field boundaries, headlands and dead furrows at the small‐watershed scale. Various time scales (yearly, daily) must also be taken into account. We then describe how a range of agricultural techniques that may reduce runoff risk on farmland can be found. In particular, we show how to analyse the farmers' leeway for adopting more opportune cultivation techniques specifying the impact of these techniques on plant growth or on work organization. But other factors such as social factors (farmers' network) also interfere and should be taken into account in soil and water protection programmes. Copyright © 2004 John Wiley & Sons, Ltd.

With the development of deep learning technology, deep reinforcement learning (DRL) has successfully built intelligent agents in sequential decision-making problems through interaction with image-based environments. However, learning from unlimited interaction is impractical and sample inefficient because training an agent requires many trial and error and numerous samples. One response to this problem is sample-efficient DRL, a research area that encourages learning effective state representations in limited interactions with image-based environments. Previous methods could effectively surpass human performance by training an RL agent using self-supervised learning and data augmentation to learn good state representations from a given interaction. However, most of the existing methods only consider similarity of image observations so that they are hard to capture semantic representations. To address these challenges, we propose spatio-temporal and action-based contrastive representation (STACoRe) learning for sample-efficient DRL. STACoRe performs two contrastive learning to learn proper state representations. One uses the agent’s actions as pseudo labels, and the other uses spatio-temporal information. In particular, when performing the action-based contrastive learning, we propose a method that automatically selects data augmentation techniques suitable for each environment for stable model training. We train the model by simultaneously optimizing an action-based contrastive loss function and spatio-temporal contrastive loss functions in an end-to-end manner. This leads to improving sample efficiency for DRL. We use 26 benchmark games in Atari 2600 whose environment interaction is limited to only 100k steps. The experimental results confirm that our method is more sample efficient than existing methods. The code is available at https://github.com/dudwojae/STACoRe.

*Philippe Martin, INA P-G, Dpt AGER, 16 rue Claude Bernard 75231 Paris cedex 05, France; François Papy, INRA SAD INRA SAD Ile de France BP 01, 78850 Thiverval Grignon, France; Alain Capillon, CIRAD direction du dept des cultures annuelles Agropolis BP 5035 34032 Montpellier cedex 1, France. *Corresponding author: pmartin@inapg.inra.fr ABSTRACT In order to characterize the effects of agricultural field state on runoff in the Pays de Caux (France), a range of actual agricultural practices has been tested under natural rainfall during two intercrop periods (1993-1994; 1994-1995). Work was conducted on 20 m2 experimental plots. Plot state descriptors included soil surface descriptors (tortuosity index in the tillage direction (TI-L), percentage of surface covered by vegetation (COV), macroporosity (MAC)) and soil profile descriptors (percentage of the anthrophic horizon tilled during intercrop period (TSW), percentage of compacted zone in the anthrophic horizon (COMP)). Climate descriptors, defined for each rainfall sequence, included cumulative rainfall (CR), mean rainfall intensity (I) and cumulative daily climatic balance (precipitation minus evapotranspiration) during the three days preceding the maximum rainfall intensity (CB3). The LIFEREG linear regression method (SAS), chosen because it allowed taking into account collector tank overflowing. The LIFEREG method, conducted on 1993-1994 results, only excluded the MAC and TSW variables (0.01 probability level). The regression equation was: Runoff (mm) = 0.419-0.200*(TI-L)0.015*COV+0.018*COMP+ 0.054*CR+0.530*I+0.018*CB3 A runoff grid composed of four classes has been set up. After all runoff sequences had been split into these four classes the regression equation has been used to determine the runoff class of each runoff sequence. The calculated class was correct for respectively 52 % of the runoff sequences in 1993-1994 and 56 % in 1994-1995. More-than-one-class error frequency only reached 3 % in 1993-1994 and 9% in 1994-1995. The equation would be useful (i) to compare different plot state at the same time and (ii) to discuss the best way to decrease runoff risk on agricultural plots.

*V. Souchere, INRA-SAD Ile de France, RN 10 (Route de Saint-Cyr), 78026 Versailles cedex (France); O. Cerdan, Y. Le Bissonnais, A. Couturier, and D. King; INRA-Science du sol, Centre de recherche d’Orléans, BP 20619, 45166 Ardon cedex (France); F. Papy, INRA-SAD Ile de France, BP 1, 78850 Thiverval Grignon (France). *Corresponding author: souchere@versailles.inra.fr ABSTRACT In loamy areas of Northern Europe, soil erosion is a widespread phenomenon, despite low rainfall intensity and a gentle topography. Interactions between meteorological conditions, farming operations and topsoil texture bring about rapid and significant changes in the hydraulic properties of topsoil. The processes involved in surface crusting are extremely dynamic and crust characteristics are often difficult to measure. Modeling infiltration into these crusts has led to the development of equations of varying complexity, ranging from simple empirical equations to numerical solutions of the Richards equation. Obtaining the parameters for the more mechanistic approaches remains a challenge. The objective of our work is to develop a simple runoff and erosion model based on field experiments and knowledge about crusting and agricultural practices (tillage direction, roughness, location of dead furrows, etc.). The model calculates the total runoff volume for a rainfall event at any point in the watershed. This model is able to serve as a simulation tool in order to test several anti-erosion schemes and choose the more efficient scheme for a given context.

This paper describes a new erosion model to predict the location and volume of ephemeral gullies within the main runoff collector network of agricultural catchments. This model, using an expert-based approach, combines field experiment results and knowledge about erosion processes and agricultural practices. It takes into account slope gradient, parameters reducing runoff flow velocity or increasing soil resistance (land use, plant cover percentage, roughness and soil surface crusting stage), the hydrological structure of catchments and the runoff volume. The model is used to calculate the soil sensitivity to ephemeral gully erosion at any point in four small cultivated catchments.Results show that it is possible to predict gully erosion from simple information that can easily be recorded by farmers. However, our model tends to overestimate the erosion level in some cases. Furthermore, the quality of the results varies strongly according to the catchment and to the rainfall event used. To increase the quality of the results, it will be necessary to improve our knowledge database from experimental results and to use a calibration procedure. (C) 2003 Elsevier Science B.V. All rights reserved.

Pour mieux orienter la recherche de solutions techniques qui limitent le ruissellement et l'érosion, les auteurs utilisent des modèles de décision dans les exploitations agricoles. En appréciant les marges de manoeuvre des agriculteurs, on évite de proposer des solutions non réalisables. Dans une région très sensible au ruissellement et à l'érosion, le Pays de Caux (Seine-Maritime, France), après enquête auprès d'agriculteurs, on a modélisé les décisions d'assolement et d'organisation du travail. On teste ainsi s'il est possible de localiser autrement les successions de cultures sur le territoire de l'exploitation, de façon à réduire le ruissellement des zones où il est le plus nocif et s'il est envisageable d'introduire dans l'organisation du travail de nouvelles techniques. Il ressort de l'application de ces modèles qu'il est impossible de reconsidérer la localisation des cultures. En revanche, on peut introduire de nouvelles techniques pendant les périodes de culture et surtout d'interculture, avec un résultat qui reste aléatoire. Les techniques qui n'ont d'autres finalités que de réduire le ruissellement ne sont évidemment pas prioritaires aux yeux de l'agriculteur lorsque leur réalisation entre en concurrence avec les objectifs de production. Plutôt que d'introduire de nouvelles techniques, il faut donc plutôt rechercher des modalités d'intervention qui rendent compatibles maîtrise du ruissellement et élaboration du rendement.

On-farm technical management of annual crops is a recurrent task, so farmers can to a large extent plan their cropping operations. Taking winter wheat on arable farms in one part of France as an example, this planning is represented in a conceptual model consisting of a set of descriptive variables and decision-making rules. Six descriptive variables and five types of decision-making rule have been identified. Three of the variables describe the intended timing of cropping operations, one determines the possible modes of operation and one groups the fields sown to wheat into sets so that different management modes can be applied to each set. We demonstrate that this model accounts for the kinds of planning practised by the farmers in our survey: taking 3 successive years' wheat crops on these farms, with sharp year-to-year differences in weather patterns, we show that planning largely determines the technical management actually exercised. The model proposed enables one to classify and understand the observed variability in farmers' practices, putting the weather in its place among other factors. This representation of decision-making processes in the form of a ‘model for action’ has been validated on wheat in other parts of France and is currently being validated for other annual crops.

Collective management of cropping systems is common in tropical countries when one production factor (e.g. machinery or an irrigated scheme) is shared by a number of farmers. To achieve their technical objectives in these contexts, farmers have to co-ordinate decision-making processes among themselves and with their economic partners. A 3-year study was carried out on two irrigated schemes in the Senegal river delta. Its aim was to understand (1) the problems farmers' organizations managing schemes faced when carrying out annual double cropping of rice and (2) how they managed to co-ordinate the different actors (individual farmers, contractors and collective organizations) interacting on these schemes. The results presented here relate mainly to the harvest part of the problem. They show double cropping success varies from one site to another and from one year to another, depending on different starting dates and global harvest performances. The comprehensive model proposed to explain this diversity includes (1) analysing individual farmers' and contractors' decision-making processes indicating the uncertainty of their behaviour and its effects on plot conditions (maturity and trafficability) and machinery performances and (2) classifying collective co-ordination processes under three main strategies of contractualizing relations with local contractors by granting harvest monopolies, simplifying complex decisions such as choosing a harvest starting date at scheme level and adjusting to unforseen events during task operation, mainly by looking for extra combine harvesters. The efficiency of these strategies is analysed in relation to the structural characteristics of the two schemes. In our discussion we propose a general framework to explain the co-ordination problems met by farmers in this context including lack of experience, diversity of individual behaviour and uncertainty. Some suggestions are put forward to improve and accelerate the organizational learning processes already acquired by farmers, in terms of technical references, scheme design and modelling.

In areas of intensive agriculture, e,g, 'Pays de Caux' in France, which was the study area, field observations have shown that runoff directions were modified by agricultural activities. In order to account for factors responsible for modifications of the runoff direction (roughness, tillage direction and agricultural patterns, e,g, dead furrow or dirt tracks), we constructed a discriminant function based on field observations, This function enables us to decide whether flow direction for slopes of up to 15% was imposed by slope direction or tillage direction. It can be applied to any location, provided there are known roughness, known slope intensity, known aspect and known tillage azimuth.In order to examine the effects of these agricultural activities at the catchment scale, we compared two models by analysing the same hydrological variables: the area contributing to runoff and the flow network. The first model (Topo) was built according to the runoff direction derived from a Digital Elevation Model (DEM). The second model (Tillage) was constructed by combining information from the DEM, and information from rules based on field observations or resulting from statistical analysis.For 23 basic catchments, the result of the comparison between the two models (Topo and Tillage) showed that a major part of the catchments and the drainage network was affected by modifications related to the introduction of man-made agricultural factors. For example, for 20 of 23 catchments, the runoff flows over more than 50% of the surface of such areas were produced along the direction imposed by tillage, The introduction of tillage effect brings about modifications of both the shape and size of catchments, (C) 1998 Elsevier Science B.V. All rights reserved.

Despite a generally low intense pluviometry and a moderate relief, the North of France is affected since about two decades by important erosive phenomenons. They are due to the soil waterproofing under the action of rainfalls and agricultural activities.The aim of this work is to show the role of the spatial distribution of runoff contributing areas within catchments in the runoff generation and concentration at the origin of erosive phenomenons. This spatial distribution changes during time according to dates and places of agricultural works and also according to meteorological conditions. In order to give an account of spatial and temporal interactions between factors, we have developed a model of erosion which attempts to predict water flow everywhere in the catchment during an agricultural year.The survey was conducted on thirty zero order catchments which were studied during two agricultural years (91-92 and 92-93). Results show that taking into account the degradation of soil surface, the arrangement of runoff contributing area and the connection between these areas allows to locate a gully in space and in time and to predict the importance of the incision.

This is an overview of the studies conducted in France (mainly in the SAD department of Inra) in response to the long established criticism of the descending linear model for the diffusion of innovation. The field addressed is restricted to the management of technical systems of production. An analysis of practices was undertaken to understand " why farmers do what they do " and the team at the SAD department then used decison-making rules to construct models. The term action model is used to describe the anticipated organisation of decisions, which consists of conceptually defining blocks of time and space so as to handle the uncertainty facing the farmer in a hierarchical way. These action models are not explicit, they must be constructed in collaboration with the farmer. This approach reveals the farmers knowledge, which can be the result of the farmers experience or inteactions with other farmers and technical experts.These studies are designed to provide decision support, and to organise advice. They require new relationships between fanners and their advisors and have implications for the approach to be used by the advisors advisors and have implications for the approach to be used by the advisors themselves. By implicating technical advisors in their work, the researcher have to develop new tools for advice, for both individual and group consulting. These objectives are particularly appropriate when a new common agricultural policy (CAP) is being implemented.

This study analyses variations in rill erosion as a function of the morphological, pedological and land use characteristics of cultivated catchments. 20 elementary catchments in Northern France, from 3.7 to 100 ha, were studied during the winter of 1988/89. The between-catchment differences in total and talweg rill volumes (m3) and rates (m3/ha) were very large, and could not be explained by the variations in rainfall. The size of runoff contributing area, defined by combining soil susceptibility to crusting and certain land use characteristics, was found to be the main factor accounting for variations in erosion. Other factors, such as the soil sand content, talweg incisable length, catchment compacity index and proportion of upslope contributing areas were also correlated with total and talweg rill volumes, but slope characteristics were not. Cropping systems greatly affected the catchment susceptibility to rill erosion through their influence on runoff generation, runoff concentration and soil susceptibility to scouring.

In the arable regions of north-western Europe, some morphological sites can generate runoff and thus cause catastrophic flooding. The paper addresses the following questions: does soil surface state influence these exceptional events as it does for more frequent events? In the Pays de Caux (near the mouth of the Seine), a particularly sensitive region, an inventory of catastrophic flooding, based on events recorded in a local weekly, has been made over a 30-yr period. These events are presented according to their occurrence throughout the year (fig 2), their chronology over the 30-yr period (fig 3) and their spatial repartition (fig 2). Two kinds of catastrophe can be distinguished: 1), winter floods which result from long rainy periods (table III) and need large watersheds (table I); those which cause the most damage are floods due to an exceptionally long and rainy period during wheat sowing; after that time, they are late enough to allow the soil surface to become crusted (table VI); 2), the flash flood caused by spring storms do not need large watersheds (table I). They are correlated both with storm frequency in spring and the amount of winter rain (table IV). However, the correlation of these catastrophes with spring storms is not perfect in that the highest frequency of spring floods appears in June but storms are equally likely to occur in May and in June (fig 4); this is interpreted as an effect of soil surface state. Large amounts of winter rain have a tendency to create rills which in turn increase spring catastrophes by reducing the runoff concentration time. The chronology of flash flood over the 30 yr (fig 3) can be explained by climatic conditions but also by agricultural land use. With agricultural inventories (table II), it is possible to induce soil surface states in January and May (table V) and thus represent, for a long period, the evolution of soil surface state in these 2 months (fig 5). It would seem that the floodings in the last 10 yr were caused by the increase in streaming surfaces. So the distribution of catastrophes throughout the year as well as over a 30-yr period show the effect of soil surface state of agricultural land on flooding. Thus flood control should be planned taking into consideration the soil surface state of the agricultural area.

On a cherché à mettre en évidence les conditions d'apparition des rigoles et ravines temporaires typiques de l'érosion par ruissellement concentré qui prévaut dans les régions limoneuses du Nord-Ouest de l'Europe.Pour cela, on a observé pendant 3 campagnes, le comportement de 9 petits bassins-versants du Pays de Caux.L'apparition et l'aggravation des figures d'érosion sont mises en relation avec les caractéristiques des pluies et avec les états de surface des parcelles constituant l'impluvium.Il apparaît que la probabilité d'érosion dépend étroitement du degré de développement des croûtes de battance, de la rugosité de surface, et de la présence d'empreintes de roues à la surface du sol.Lorsque l'érosion est provoquée par le ruissellement concentré, c'est donc sur la base de ces critères qu'il faut apprécier l'influence des systèmes de culture.En effet, ceux-ci induisent une forte variabilité spatiale et chronologique des propriétés hydrauliques du terrain.