Introduction Directory UMM :Data Elmu:jurnal:A:Agriculture, Ecosystems and Environment:Vol81.Issue1.Oct2000:

Agriculture, Ecosystems and Environment 81 2000 5–16 A spatial approach using imprecise soil data for modelling crop yields over vast areas Philippe Lagacherie a,∗ , Durk R. Cazemier a , Roger Martin-Clouaire b , Tom Wassenaar a a INRA Science du Sol, 2 Place Pierre Viala, 34060 Montpellier Cedex, France b INRA, Biométrie et Intelligence Artificielle, BP27, 31326 Castanet-Tolosan Cedex, France Received 25 August 1999; received in revised form 10 December 1999; accepted 28 February 2000 Abstract Estimations of crop yields using process-based crop models are area-limited because quantitative soil data are unavailable over vast areas. The spatial approach proposed in this study incorporates two novel aspects concerning the derivation of soil data feeding the simulation and the modelling of the crop production process. First, the soil parameters required for crop modelling as well as their imprecision were estimated from the qualitative information of a 1:250,000 scale regional soil database by a possibility theory approach, which combines a set of GIS procedures and a constraint satisfaction solver. Second, the initial process-based crop model was made less complex by deriving simple agrotransfer functions from simulations at representative sites located in the studied region. The resulting system estimated the yield expressed as possibility distributions over the region which can be visualised through decision maps. The proposed spatial approach was tested on a hard wheat yield Triticum durum spp. estimation in the Hérault-Orb-Libron valley region Languedoc, France. It proved to provide realistic yet imprecise estimates compared with those obtained for a set of site-crop estimates. The spatial approach allowed the identification of areas in which soil data needed to be improved for obtaining both reliable and informative estimates. These results demonstrated the potential usefulness of the proposed approach for providing reliable soil information for decision making at the regional level. © 2000 Elsevier Science B.V. All rights reserved. Keywords: Soil map; Available water capacity; Regional scale; Crop model; Imprecision; Possibility theory; Constraint satisfaction problem; GIS

1. Introduction

Computer simulations of soil water regimes and crop growth are powerful means for quantifying the effects of changing climate conditions or agricultural practices. However, these simulations require a large amount of quantitative soil data which limits their spa- tial application to areas where a dense spatial sampling ∗ Corresponding author. Tel.: +33-4-99-61-25-78; fax: +33-4-67-63-26-14. can be undertaken. A spatial approach is needed for extending the simulations to vast areas such as Euro- pean regions. The common practice for doing this consists in linking crop models to quantitative information pro- vided by soil maps and soil databases Dumanski et al., 1993; Smaling and Fresco, 1993; Akinremi et al., 1997; Bornand et al., 1998. This type of informa- tion consists in measured soil data from detailed soil profile observations that are assumed to be represen- tative for the delineated mapping units. Crop yield 0167-880900 – see front matter © 2000 Elsevier Science B.V. All rights reserved. PII: S 0 1 6 7 - 8 8 0 9 0 0 0 0 1 6 4 - X 6 P. Lagacherie et al. Agriculture, Ecosystems and Environment 81 2000 5–16 simulations can then be undertaken by using these input data, if necessary, in combination with pedo- transfer functions Bouma and van Lanen, 1987. The so obtained results are assumed to be valid for the whole mapping unit, thus allowing easy mapping by classical GIS procedures. This straightforward approach becomes question- able when yield must be mapped over large areas. Generally, small scale soil maps 1:250,000 are the only available soil information over such areas. At this scale the mapping units often include several taxonomic units, each of them characterised by a rep- resentative profile. The common practice is to select the representative profile of the dominant taxonomic units which can lead to neglect a substantial part of the mapping unit area Le Bas et al., 1998. Further- more the taxonomic units of a small scale soil map cannot be reduced to the description of a unique set of parameters measured at a given site because of its high within-unit variability. Resuming, the exclusive use of quantitative information derived from a small scale soil map leads to a misrepresentation of the soil variability of the region. This may have consequences on yield mapping and further decision making. A possible alternative is to use the qualitative de- scription of the soil taxonomic units STUs of a small scale map as input data for crop models. This description provides a more synthetic understanding of the taxonomic units taking into account the entire information collected by the soil surveyor in the field, i.e. soil profiles, auger hole observations and surface features observations. The description of a STU also includes a description of environmental attributes e.g. geology, land use, slope. If maps of these at- tributes exist for the studied region, the environmental descriptions can be used for estimating the location of each STU within the complex soil mapping units of the 1:250,000 scale soil map. All this qualitative information, generally based on well-established codification systems FAO-UNESCO, 1981; Baize and Jabiol, 1995, is now easily avail- able in soil databases Oldeman and van Engelen, 1993; Bornand et al., 1994; King et al., 1994. This information is, however, still underexploited in yield predictions. The approach presented below estimates crop yields over a region by coupling simulations with the qualitative descriptions of taxonomic soil units of a 1:250,000 soil database. Such coupling is made through agrotransfer functions derived from a set of crop model simulations undertaken over a set of representative soil climate situations. The proposed approach uses possibility theory for representing and handling the required qualitative information. It was applied within the framework of the EC funded re- search project IMPEL Rounsevell et al., 1998 for mapping hard wheat yields evolutions with changing climate conditions over a region of 1200 km 2 located in the Languedoc plain southern France. This paper focuses on the estimation of wheat yield under actual climate conditions. The climate changes issues dealt in the IMPEL project are detailed in Wassenaar et al. 1999.

2. The spatial approach