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

Agriculture, Ecosystems and Environment 81 2000 57–69 Adjustment procedures of a crop model to the site specific characteristics of soil and crop using remote sensing data assimilation M. Guérif ∗ , C.L. Duke 1 Unité d’Agronomie I.N.R.A. de Laon Péronne, rue F. Christ, 02007 Laon Cedex, France Received 14 July 1999; received in revised form 27 January 2000; accepted 14 February 2000 Abstract Crop models can be useful tools for estimating crop growth status and yield on large spatial domains if their parameters and initial conditions values can be known for each point. By coupling a radiative transfer model with the crop model through a canopy structure variable like LAI, it is possible to assimilate, for each point of the spatial domain, remote sensing variables like reflectance in the visible and near-infrared, or their combination into a vegetation index, and to re-estimate for this point some of the parameters or initial conditions of the model. The spatial adjustment of the crop model provides, therefore, a better estimation of yield. This process has been tested on the re-estimation of crop stand establishment parameters and initial conditions for sugar beet Beta vulgaris L. crops, using the crop model SUCROS coupled to the radiative transfer model SAIL. The quality of recalibration depends in particular on the precision on the SAIL parameters other than LAI: soil reflectance, optical properties of leaves, and leaf angles. In applications on large domains, where these parameters may vary a lot, their estimation is an important factor in recalibration error. It was shown, using stochastic simulation, that beforehand knowledge of the variability of soil and crop characteristics considerably improved the results of the assimilation of reflectance measurements. The best results were obtained when the spectral reflectances were combined into the TSAVI vegetation index, which minimised the disruptive contribution of soil to the canopy reflectance. In particular, the use of TSAVI provided more consistent results for the estimates of the sowing date and emergence parameters, which remained poorer than yield estimates. Over a wide range of unknown crop situations corresponding to extreme sowing dates and emergence conditions, the proposed method allowed to estimate the sugar beet yield with relative errors varying from 0.6 to 2.6. The results appeared to depend on the timing of remote sensing data acquisition; the best situation was when the data covered the whole period of LAI growth including the highest values. © 2000 Published by Elsevier Science B.V. Keywords: Local calibration; Reflectance; TSAVI; Error propagation; Stochastic simulation; Beta vulgaris L. ∗ Corresponding author. Tel.: +33-3-23-23-64-88; fax: +33-3-23-79-36-15. E-mail address: martine.gueriflaon.inra.fr M. Gu´erif 1 Present address: Land Resource Research, Guelph University, Guelph, Ont., Canada N1G 2W1.

1. Introduction

Crop models can be useful tools for estimating crop growth and yield on large spatial domains in order to monitor storage capacities, factories supply and or- ganisation. One of the greatest difficulties is that there is a need to know the value of parameters andor initial 0167-880900 – see front matter © 2000 Published by Elsevier Science B.V. PII: S 0 1 6 7 - 8 8 0 9 0 0 0 0 1 6 8 - 7 58 M. Gu´erif, C.L. Duke Agriculture, Ecosystems and Environment 81 2000 57–69 conditions for each field in order to apply the model to each point of the domain, and these may vary con- siderably from one point to another. This information is generally not available, but remote sensing, espe- cially in the optical domain, which gives extensive spatial information on the real crop growth status, is a practical way of estimating it. Several methods have been explored Delécolle et al., 1992. One of them Bouman and Goudriaan, 1990; Guérif and Duke, 1998 consists of coupling a radiative transfer model to the crop model through a canopy structure variable like LAI, which makes it possible to simulate remote sensing variables like reflectance in the visible and near-infrared for each point of the spatial domain, at the times when remotely sensed reflectance data are available. Comparison of the simulated and measured variables allows re-estimation of some of the parame- ters or initial conditions of the model, leading to better simulation of yield. This process is called the assim- ilation of remote sensing data into the crop model; it provides a local adjustment of the crop model. Such a procedure was developed in a previous study on sugar beet Guérif and Duke, 1998 that employed the SUCROS model Spitters et al., 1989 and the SAIL model Verhoef, 1984 for simulating radiative transfer into the canopy. The parameters describing the establishment of the crop time between sowing and emergence, number of plants emerged, initial leaf area, which vary greatly according to soil, weather and sowing techniques were re-estimated for specific situations. However, there is a need to know the value of the parameters of the radiative transfer model other than LAI, which are specific to the status of canopies and soils optical properties and geometry of the leaves, reflectance of the soil, and so vary greatly from place to place over a large domain. One should think about estimating them within the assimilation of remote sensing data technique also, but it would increase the number of parameters to be estimated and require a high amount of remote sensing data which is hardly available in operational situations. It was shown that it is possible to develop a method for estimating the values of these parameters, based on prior analysis of their regional variations Duke and Guérif, 1998. Using this method, instead of standard values, can re- duce the errors in spectral reflectance estimates made by the radiative transfer model. The objective of this paper is to measure the in- fluence of these remaining errors due to uncertainty of soil and canopy parameters on the result of re- mote sensing data assimilation process, and hence on the estimates of crop model parameters, initial conditions and yield. The overall performance of the method was evaluated for estimating crop parameters and yield in virtual regional conditions where neither the initial condition sowing date nor some important crop parameters crop establishment characteristics are known. The study is restricted to sugar beet crops in northern France Picardie. In the first part of the paper, the main statements of the previous mentioned studies are presented, then the methodology used and the results obtained in this work of error propagation analysis are presented and discussed.

2. The procedure used for crop model site specific adjustment