JXB Advance Access originally published online on January 13, 2007
Journal of Experimental Botany 2007 58(4):869-880; doi:10.1093/jxb/erl231
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Remote Sensing |
Quantification of plant stress using remote sensing observations and crop models: the case of nitrogen management
INRA-CSE, Site Agroparc, F-84914 Avignon, France
* To whom correspondence should be addressed. E-mail: baret{at}avignon.inra.fr
Received 14 June 2006; Accepted 16 October 2006
 |
Abstract |
|---|
 Top Abstract IntroductionEstimation of canopy state...Linking remote sensing estimates...Using crop models for...ConclusionReferences |
|---|
Remote sensing techniques offer a unique solution for mapping stress and monitoring its time-course. This article reviews the main issues to be addressed for quantifying stress level from remote sensing observations, and to mitigate its impact on crop production by managing cultural practices. The case of nitrogen fertilization is used here as a paradigm. The derivation of canopy state variables such as the leaf area index (LAI) and chlorophyll content (Cab) is first addressed. It is demonstrated that the inversion of radiative transfer models leads to useful estimates of these variables. However, because of the ill-posed nature of the inverse problem, better accuracy is achieved when using prior information on the distribution of the variables and when multiplying LAI by Cab to get canopy level chlorophyll content. This variable, LAIxCab is well suited for quantifying canopy level nitrogen content. It is used for nitrogen stress evaluation by comparison with a reference unstressed situation which is, however, not easy to get in practice. The combination of remote sensing observations with crop models provides an elegant solution for stress quantification through assimilation approaches. It fuses several sources of information within our knowledge of the processes involved and accounts for the environmental budget which can be integrated when making decisions about cultural practices. Conclusions are drawn on the issues related to the retrieval of canopy state variables from remote sensing data, to the link between these observables and crop models, and to the assimilation approaches. Avenues for further research are finally discussed along with the required observation system.
Key words: Chlorophyll, functioning model, inversion, nitrogen, precision farming
|
Introduction |
|---|
|
TopAbstractIntroductionEstimation of canopy state...Linking remote sensing estimates...Using crop models for...ConclusionReferences |
|---|
Stresses come from a range of factors that limit the potential growth of canopies. Evaluation of the stress level to which plants are subjected is therefore critical information required both for the quantification of consequences on production and for taking action for their mitigation. Many stresses are caused directly or indirectly by water shortages or insufficient satisfaction of canopy mineral requirements, particularly nitrogen. Fungi and pests may also induce stresses by reducing the green leaf area, the sap flow, or by altering plant metabolism. Therefore, most stresses show-up with similar symptoms: (i) leaf area index (LAI) reduction, due both to a decrease of leaf area production and to an increased senescence rate; (ii) chlorophyll content (Cab) decrease due either to a limitation of the synthesis of chloroplasts, to their destruction, or to the remobilization of this important protein pool; (iii) canopy temperature increase due to lower transpiration fluxes. Stress will therefore be expressed by changes in a few canopy state variables such as LAI, chlorophyll content, or surface temperature. However, clear identification of the stress factors is not always possible without using additional information.
The signal emitted or reflected by vegetation surfaces is governed by state variables describing canopy structure (leaf area and orientation, spatial arrangement, roughness) and on the optical, dielectric, or thermal characteristics of the vegetation elements. The physical quantity recorded by remote sensors depends on the spectral domain considered where different physical processes may be involved: (i) in visible, near and short-wave infrared domains, the physical quantity considered is reflectance, i.e. the fraction of radiation reflected by the surface; (ii) in thermal infrared, canopy is characterized by its brightness temperature corresponding to the radiation flux emitted; (iii) in active microwaves, the back-scattering coefficient is measured by radar systems; (iv) in passive microwaves, the canopy is characterized similarly to the thermal infrared domain by its brightness temperature where however, emissivity plays a key role.
Each spectral domain is therefore sensitive to specific canopy state variables (Table 1). Canopy structure variables are mainly accessible in the solar reflective domain, i.e. from the visible to the short-wave infrared. Microwaves are sensitive to water content (Paloscia et al., 1999; Prévot et al., 2003; Wigneron et al., 2004). Chlorophyll and water content in the leaf may be estimated with a reasonable accuracy from the solar reflective domain (Combal et al., 2002; Ustin et al., 2004) because of strong and specific light absorption bands. The thermal infrared domain provides information on leaf surface temperature governed by the energy and water balance of the canopy (Olioso et al., 2002). For stress detection and quantification, the reflective solar domain appears very pertinent, particularly regarding the capacity to estimate leaf area index (LAI) and chlorophyll content (Cab). Both variables are widely used by agronomists and ecologists to diagnose possible stress affecting the canopies. This is the basis for applications to agricultural management techniques developed recently, such as precision farming for which the spatial heterogeneity within and between fields is explicitly accounted for through site-specific practices (Cook and Bramley, 1998; Moran et al., 1997; Pierce and Nowak, 1999). Remote systems sample the radiance field reflected or emitted by canopies at spatial resolutions and frequencies suitable for such applications, i.e. from sub-metric to metric and deca-metric pixel size, allowing mapping the spatial heterogeneity and the associated stress levels. Sensors in the visible and near infrared spectral domains are available aboard a range of vectors, from satellite, airborne, balloons, tractors, down to hand-held devices. They sample the temporal dimension over a range of frequencies, although cloud occurrence is a serious limitation for observations in the solar domain from high altitude vectors.
|
Deriving stress maps from the observed radiance field is a complex process (Guérif et al., 2001). The radiance field should first be translated into canopy state variable maps such as LAI or chlorophyll content. Then the stress level quantification at each point of the map may be achieved by comparison with unstressed canopies placed in similar conditions. This assumes that unstressed areas are known and precisely located in the fields, and representative of all the field conditions apart from the targeted stress, which is obviously difficult to organize. Other references are thus required, and crop models may provide a good alternative: they allow accounting for a number of forcing variables such as climate, soil or cultural practices. They also allow quantifying the stress level and its impact on the final production. However, models are not perfect and require a large number of parameters and initial conditions not always very well known and that may vary spatially. Combining radiometric information with crop models within an assimilation approach is very appealing for stress quantification. In addition, crop models may be used in a prognostic mode to evaluate several scenarios of cultural practices under variable climatic conditions and to select the optimal one that maximizes the economic output for the farmer, while preserving the environment.
This article reviews the several steps required to go from remote sensing observations to stress quantification and making optimal decisions in agriculture. The approach is illustrated by nitrogen stress which is one of the main drivers of yield for many species. Although the unit price of nitrogen fertilizer is generally low regarding the possible impact on yield and grain quality, excess nitrogen may strongly impact the environment through leaching in the water table, and may also lead to lodging and increased vulnerability to several diseases. It is therefore pertinent to develop methods allowing sustainable nitrogen fertilization management.
The problems associated with the derivation of canopy state variables from remote sensing observations are addressed first. Then several ways to link these canopy state variables to stress levels are described. Finally, the combination of remote sensing observations and crop models is investigated, with the application to stress quantification and the optimal control of cultural practices.
|
Estimation of canopy state variables from remote sensing observations |
|---|
|
TopAbstractIntroductionEstimation of canopy state...Linking remote sensing estimates...Using crop models for...ConclusionReferences |
|---|
The spectral and directional distribution of top of canopy reflectance at a given time and location is governed by the canopy structure, optical properties of the elements including the soil background, as well as view and illumination geometrical configuration. Because of numerous variables influencing canopy radiometric response, it is not straightforward to extract specific canopy variables. Radiative transfer models simulate the reflectance field as a function of the whole set of influencing variables including LAI and Cab. However, retrieving these variables of interest from measurements of top of canopy reflectance as sampled with a particular sensor, requires inverting radiative transfer models.
Figure 1 presents the general scheme of canopy variable retrieval from top of canopy (TOC) remote sensing observations. A radiative transfer model simulates in the forward direction the TOC reflectance field from the input variables. These were split into the variables of interest such as LAI and Cab and the other characteristics (geometrical view and illumination configuration, other canopy variables such as leaf inclination, leaf water content, soil optical properties...). An inverse technique is then used to extract the variables of interest. It generally requires prior information on the distribution of the input variables to regularize the inversion process (Tarentola, 1987).
|
As a matter of fact, several combinations of input variables may provide very similar reflectance simulations that match closely the actual remote sensing observations as illustrated by Fig. 2. This corresponds to the generally ill-posed nature of the inverse problem in remote sensing (Combal et al., 2002). The radiometric information is not sufficient to identify a unique solution: the inverse problem needs to be regularized by exploiting additional information such as prior knowledge on the statistical distribution of canopy radiative transfer model input variables. Closer inspection of the possible solutions (Fig. 2) shows a strong negative correlation between the retrieved LAI and Cab values. All the solutions have very similar values of canopy integrated chlorophyll content corresponding to the LAIxCab product between leaf level chlorophyll content (Cab) and LAI. Note that canopy integrated chlorophyll content is a physically sound quantity since it represents the optical path in the canopy where absorption by chlorophyll governs the radiometric signal. Note also that canopy integrated chlorophyll content neglects chlorophyll contained in other elements than leaves such as stems.
|
When prior information on canopy and soil input variables are exploited within the inversion process, improvement of the retrieval performances is observed (Fig. 3). The accuracy and robustness of canopy characteristics estimation is even improved when using canopy level chlorophyll content (LAIxCab) as compared to leaf level contents (Cab). In the following, the interest in using canopy integrated chlorophyll content variable will be further demonstrated with reference to its role as a pertinent diagnostic variable.
|
|
Linking remote sensing estimates of canopy variables to stress diagnostic variables: the case of nitrogen stress |
|---|
|
TopAbstractIntroductionEstimation of canopy state...Linking remote sensing estimates...Using crop models for...ConclusionReferences |
|---|
The first study conducted to evaluate the use of remote sensing for nitrogen stress quantification were mainly based on empirical relationships with spectral indices that were sensitive to chlorophyll content (Penuelas et al., 1994). Other authors (Bausch and Duke, 1996) use leaf chlorophyll content estimated with the SPAD instrument (Ercoli et al., 1993) to evaluate canopy nitrogen status with reference to a non-stressed canopy. More recently, Vouillot et al. (1998) directly relate the nitrogen nutrition index as defined by Justes et al. (1994) and Lemaire et al. (1989) to the ratio between red and near infrared reflectance. However, as demonstrated previously, canopy radiometric response depends on many variables, and there is little chance that such a simple ratio between two bands shows robust relationships with some of the biophysical variables of the canopy as demonstrated by several other studies (Baret and Guyot, 1991). Nevertheless, the concept of optimal nitrogen concentration proves pertinent since it constitutes the reference case required to identify nitrogen stress and even to quantify it. Nitrogen nutrition index (NNI) is defined with regard to the optimal nitrogen concentration, NR, achieving maximum biomass production (Fig. 4). NR values depend on the biomass amount according to the dilution theory (Lemaire and Gastal, 1997):
 |
|
The nitrogen nutrition index is simply the ratio between the actual nitrogen concentration, N, to the optimal value, NR: NNI=N/NR. If N >NR, then NNI >1.0 and the canopy will not use efficiently the nitrogen amount which is in excess. Conversely, a canopy is assumed to be stressed if the nitrogen concentration is below the optimal value, i.e. NNI <1.0. The amount of nitrogen required by the canopy, assuming a nitrogen use efficiency of 1.0, will be
QN=W(N), where W is the actual canopy biomass and N=NR–N. It would therefore be more useful to express the optimal nitrogen level in terms of nitrogen content (kg ha–1) rather than in terms of concentration (%). Figure 4, right, illustrates the concept of optimal nitrogen content, QNR, with QNR=WNR. Agronomists have been focusing mostly on nitrogen content or concentration measurements to diagnose possible stresses. However, Baret and Fourty (1997) demonstrated that there were very few chances to retrieve canopy nitrogen directly or protein content from remote sensing observations, even using hyper-spectral systems. Nitrogen status could only be accessed through chlorophyll estimates which can actually be retrieved with reasonable accuracy as seen earlier in this paper. A dedicated experiment conducted in Laon in 2000 and 2001 over wheat crops subjected to a range of nitrogen stresses clearly demonstrated the interest of relating nitrogen to chlorophyll at canopy level rather than at leaf level (Houlès et al., 2006). The relationships obtained at canopy level between LAIxCab and QN are more robust among years and development stages than the relations obtained between N and Cab (Fig. 5). However, these relationships depend slightly on development stages: after earing, plants show larger amounts of nitrogen for the same chlorophyll amount when a large part of nitrogen is concentrated in the grains (Fig. 5). The younger stages also have more nitrogen for the same amount of chlorophyll and this might be due similarly to the dilution of nitrogen in the structural biomass with leaf development as confirmed by the strong relationship between leaf nitrogen and leaf chlorophyll content per unit soil area (Fig. 6). This clearly shows the complexity of the processes associated with nitrogen and chlorophyll distribution in plants.
|
|
Canopy integrated chlorophyll content therefore offers great potential for detecting and quantifying nitrogen stresses. The dilution curve is used here as a reference to evaluate a possible deficit in nitrogen availability when the actual biomass amount is known. This biomass can be derived from remote sensing measurements either through empirical relationships established during crop development with LAI, or by using simple light use efficiency models (Ruimy et al., 1994) where the fraction of PAR absorbed by the canopy could be derived from remote sensing observations (Weiss et al., 2000). However, the stability of these empirical relationships, or that of the light use efficiency is questionable when plants are subjected to non-optimal conditions with regard to water balance or to unusual climatic conditions, or when considering new cultivars. Using crop models accounting for all these effects may thus constitute an elegant alternative solution for setting the reference case required for stress level evaluation. Crop models may also simulate nitrogen use efficiency that may vary widely depending on climatic and soil conditions. Remote sensing observations will then provide a way to correct the trajectory of crop models when all the initial conditions or soil and canopy parameters are not perfectly known.
|
Using crop models for stress management |
|---|
|
TopAbstractIntroductionEstimation of canopy state...Linking remote sensing estimates...Using crop models for...ConclusionReferences |
|---|
Several authors have proposed using crop models as diagnostic tools (Acock and Pachepsky, 1997; Engel, 1997; Matthews and Blackmore, 1997; Bouma et al., 1999; Kropff et al., 2001; Batchelor et al., 2002; Jones et al., 2003). However, crop models require a relatively large number of input variables, initial conditions, and parameters. They may also suffer from structural deficiencies (Boote et al., 1996). They therefore have difficulties in operational applications where only a fraction of these required variables and parameters are precisely known. This is particularly true when describing the spatial variability within fields, and crop models generally show poor performances (Hayes and Privette, 1998). However, remote sensing observations can be used to keep models on track as suggested by Jongschaap (2006). Such assimilation approaches are emerging recently due to the combination of advances in remote sensing data and interpretation methods, more realistic crop models, and large improvement in computation performance (Guérif and Duke, 1998; Nouvellon et al., 2001; Bach and Mauser, 2003; Launay and Guérif, 2003; Prevot et al., 2003). These models, once recalibrated using remote sensing observations could be used in prognostic mode to select the best strategies for crop management (Moore and Tyndale-Biscoe, 1999; Houlès et al., 2004).
In the following, we will illustrate how to exploit a crop model to quantify nitrogen stresses and will propose an optimal strategy for fertilizer application in the context of precision agriculture (Houlès, 2004). The STICS model is used here to simulate canopy functioning as a function of climatic, soil, and cultural practices (Brisson et al., 1998, 2002). The approach is divided into two steps. (i) Recalibration of the model inputs and parameters using remote sensing observations. This corresponds to the assimilation step. (ii) Selection of the optimal scenario by running in prognostic mode the recalibrated model. This corresponds to the control step.
Assimilation of remote sensing observations
In the context of precision agriculture, the main issue consists in assessing the within-field variability of the main driving soil and crop variables and parameters. Therefore, a sensitivity analysis (Ruget et al., 2002) was conducted to select the most important ones that need calibration. The other variables and parameters were assessed from expert knowledge. Cultural practices and climatic variables measured during the experiments were used as model inputs. Recalibration of crop models is an inverse problem which is difficult to use, similar to that of the canopy state variables retrieval from remote sensing. The GLUE Bayesian method was therefore used (Beven and Binsley, 1992; Makowski et al., 2002), exploiting prior distribution of the variables and parameters to be estimated to get the optimal values for their later distribution. This method is summarized into the following four steps (for more details, see Houlès et al., 2004; Guérif et al., 2006) (Fig. 7). (i) A large number of parameter vectors (200 000) are randomly drawn within their prior distribution defined by expert knowledge and previous experiments. Note that at this stage, very little information is available on within-field spatial distribution of the parameters. (ii) The output variables corresponding to LAI and QN observations, derived previously from remote sensing observations, are simulated for each parameter vector using the STICS model. (iii) Observed LAI and QN values are then compared with the simulated values, and the likelihood of each parameter vector is computed. (iv) Finally, the later distribution of each parameter is estimated using the weights attributed to each parameter vector as derived from its corresponding likelihood.
|
The application of this technique is illustrated here over a wheat experiment (Houlès, 2004) where LAI and LAIxCab were estimated from airborne remote sensing measurements during four flights covering the growing season (Moulin et al., 2003). QN values were then derived from remote estimates of LAIxCab based on empirical relationships. The technique was applied to all the pixels (20 mx20 m) of the field. When using only prior information on the unknown variables, results show no spatial structure of LAI, QN, and yield at each of the four dates (Fig. 8, left), because prior distribution of the unknowns was only statistical, without incorporating the lack of knowledge on their spatial structure. Conversely, when remote sensing estimates of LAI and QN are assimilated within STICS, the spatial structure appears clearly (Fig. 8, right). The agreement between measured and simulated LAI and QN values improves significantly, the points being roughly aligned along the 1:1 line. This result was expected for LAI and QN values, since the assimilation consists in tuning the unknown soil and cultivar parameters to get the best match between STICS simulated LAI and QN values, and those derived from remote sensing. The RMSE value for LAI decreases from 1.04 to 0.57 (Fig. 8, top) after assimilation. This is not the case for QN (Fig. 8, top) where an increase in the RMSE value (from 33.4 to 40.2) is observed after assimilation. This may indicate that either the observations, the formalism of the model or the prior information used are not correct. This is more relevant for QN, where the simulated values after assimilation are significantly smaller than remote sensing estimates of QN for the higher QN values corresponding to the two latest flights. Conversely, the two first flights (smaller QN values) show good consistency with remote sensing QN values. However, the decrease of RMSE values observed for yield (from 1.81 to 0.73, Fig. 8, bottom) shows that the assimilation was efficient for this key crop model output. Inspection of the standard deviations (the vertical bars in Fig. 8) shows also that the assimilation of remote sensing observations decreases the standard deviation in some cases (QN, first flights, yield, low values).
|
Control of cultural practices
The control of cultural practices will be illustrated here for the third nitrogen application over the previous wheat experiment for which posterior distribution of the unknown parameters were adjusted as described above using the four remote sensing flights. To optimize nitrogen application over each part of the field, eight scenarios of nitrogen amount were compared (Fig. 9). Because of the unknown weather conditions between the time of the decision and harvest, 30 climate scenarios were considered using climatic records from previous years. The simulations were then performed for all the combinations of climate and nitrogen scenarios for each of the 250 pixels of 20x20 m2 covering the 10 ha of the field. This resulted in 8x30x250=60 000 simulations. Note that at this level, the later distribution of each unknown parameter was simply approximated by its median value. Athough the whole distribution was available from the previous ‘assimilation’ step, using the median value reduces drastically the number of simulations and saves a very significant amount of computer time, at the expense of a slight degradation of the accuracy of the simulations.
|
The STICS model simulated the corresponding yield (Y) and grain protein content (Ng). These outputs were used to compute the corresponding gross margin, Gm:
 |
This process applied over all the pixels of the field provides the optimal nitrogen application map as illustrated by Fig. 10. Note that no nitrogen application was required over about 20% of the field area, because canopy requirements are small enough to be covered by soil contribution. An homogeneous nitrogen application would lead to excess nitrogen left in the soil with some possible leaching into the water table. Conversely, applying a unique average nitrogen rate to the areas requiring the larger nitrogen amounts would reduce the yield and the farmer's income.
|
|
Conclusion |
|---|
|
TopAbstractIntroductionEstimation of canopy state...Linking remote sensing estimates...Using crop models for...ConclusionReferences |
|---|
Nitrogen is used here as a paradigm for stress management exploiting remote sensing observations. Most methods dedicated to nitrogen management are based on some measurements or estimates of canopy nitrogen content to decide how much nitrogen is to be applied at each location in the field. This canopy nitrogen content is compared to a reference level corresponding to an optimal nitrogen nutrition situation. The difference indicates the actual stress level and therefore the required amount of nitrogen to be applied. One of the main problems related to these techniques is the availability of such reference non-stressed situations, and the generally poorly known nitrogen use efficiency. Farmers may therefore set aside a small part of their field to apply important amounts of nitrogen to ensure unstressed reference status. However, this field sample is not necessarily representative of the other field conditions: spatial heterogeneity may be due both to variation in the soil properties, and to canopy status in relation to changes in practices (sowing depth and density, variability of previous applications ...). In addition, the amount of nitrogen required by the canopy also depends on the potential yield and nitrogen use efficiency, which vary with soil characteristics and the climatic conditions that will prevail after the nitrogen application up to the final harvest.
Crop models offer a very promising way to estimate this reference level and its variability in the field. In addition, crop models are able to account for several scenarios of climate conditions prevailing between the nitrogen application considered and harvest, frequency analysis of these variables allowing a better decision to be derived. It can also account for nitrogen use efficiency. Running crop models in such a predictive mode is very appealing for managing cultural practices as illustrated in this paper. Because models may simulate production in quantity and quality, as well as environmental variables (stock of mineral nitrogen left in the soil, for example), the decision may account for both economic and environmental constraints. It may even be used to design and tune policies dedicated to the sustainability of agricultural practices and systems.
However, the success of crop models for decision making relies on their performances for yield and environmental budget simulations. Uncertainties in model outputs depend on errors in the structure of the model and in some of its parameters and variables characterizing the initial conditions. Many of these parameters and variables vary from place to place in the field, and for most of them, there is no direct way of measuring them. The integration of remote sensing observations within such crop models would reduce these uncertainties.
Remote sensing observations in the visible and near infrared spectral domains allow canopy leaf area index and leaf chlorophyll content to be mapped. It is demonstrated here that the canopy chlorophyll content is more strongly related to the canopy nitrogen content. This provides the necessary link between remote sensing observations and canopy state variables used as indicators of nitrogen status. Fortunately, chlorophyll content is better estimated at the canopy than at the leaf levels. This is mainly explained by the possible compensations observed in the inversion process between leaf level chlorophyll content and leaf area index: several combinations of LAI and Cab may lead to very similar spectral reflectance responses. However, the uncertainties associated with canopy nitrogen content are still significant, around 20–30 kg ha–1, which is at the limit of the acceptable level. Sources of uncertainties come first from the remote sensing estimates of canopy chlorophyll content, both because of measurement uncertainties and of the lack of realism of the radiative transfer models used. In addition, the relationships between canopy and nitrogen chlorophyll contents may vary slightly depending on the situations encountered.
Further research should therefore be dedicated to the improvement of these estimates from a better description of canopy architecture in the radiative transfer models, including possible gradients in chlorophyll content, and a proper modelling of the relationships between chlorophyll and nitrogen at the canopy level. These research axes should converge towards the development of the coupling between crop models and dynamic description of canopy 3D architecture, itself coupled to a radiative transfer model. Once this whole modelling chain has been developed, remote sensing observations could be directly assimilated in terms of radiance quantities in such process models. The only remaining limiting factor will be the availability of timely observations. Technological advances now allow systems to be designed with a cluster of satellites making global and frequent observations with a spatial resolution between 5–20 m which is required for most agriculture applications.
|
References |
|---|
|
TopAbstractIntroductionEstimation of canopy state...Linking remote sensing estimates...Using crop models for...ConclusionReferences |
|---|
Acock B and Pachepsky Y. (1997) Holes in precision farming: mechanistic crop models. Precision agriculture 1997Oxford Bios Scientific Publishers Ltd pp. 397–403.
Bach H and Mauser W. (2003) Methods and examples for remote sensing data assimilation in land surface process modeling. IEEE Transactions on Geoscience and Remote Sensing 41 1629–1637.[CrossRef]
Baret F and Fourty T. (1997) Radiometric estimates of nitrogen status of leaves and canopies. In Lemaire G (Ed.). Diagnosis of nitrogen status in cropsHeidelberg, Germany Springer-Verlag pp. 201–227.
Baret F and Guyot G. (1991) Potentials and limits of vegetation indices for LAI and APAR assessment. Remote Sensing of the Environment 35 161–173.[CrossRef]
Baret F, Vintila R, Lazar C, et al. (2001) The ADAM database and its potential to investigate high temporal sampling acquisition at high spatial resolution for the monitoring of agricultural crops. Romanian Agricultural Research 16 69–80.
Batchelor WD, Basso B, Paz JO. (2002) Examples of strategies to analyse spatial and temporal yield variability using crop models. European Journal of Agronomy 18 141–158.[CrossRef]
Bausch WC and Duke HR. (1996) Remote sensing of plant nitrogen status in corn. Transactions of the ASAE 39 1869–1875.
Beven KJ and Binsley AM. (1992) The future of distributed models: model calibration and uncertainty predictions. Hydrological Processes 6 279–298.[CrossRef][Web of Science]
Boote KJ, Jones JW, Pickering NB. (1996) Potential uses and limitations of crop models. Agronomy Journal 88 704–716.
Bouma J, Stoorvogel J, Van Alphen BJ, Booltink HWG. (1999) Pedology, precision agriculture, and the changing paradigm of agricultural research. Soil Science Society of America Journal 63 1763–1768.
Brisson N, Mary B, Ripoche D, et al. (1998) STICS: a generic model for the simulation of crops and their water and nitrogen balances. I. theory and parameterization applied to wheat and corn. Agronomie 18 311–346.
Brisson N, Ruget F, Gate P, et al. (2002) STICS: a generic model for simulating crops and their water and nitrogen balances. II. Model validation for wheat and maize. Agronomie 22 69–92.[CrossRef][Web of Science]
Combal B, Baret F, Weiss M. (2001) Improving canopy variables estimation from remote sensing data by exploiting ancillary information. Case study on sugar beet canopies. Agronomie 22 2–15.
Combal B, Baret F, Weiss M, Trubuil A, Macé D, Pragnère A, Myneni R, Knyazikhin Y, Wang L. (2002) Retrieval of canopy biophysical variables from bi-directional reflectance data. Using prior information to solve the ill-posed inverse problem. Remote Sensing of Environment 84 1–15.[CrossRef]
Cook SE and Bramley GV. (1998) Precision agriculture: opportunities, benefits and pitfalls of site-specific crop management in Australia. Australian Journal of Experimental Agriculture 38 753–763.[CrossRef]
Engel T. (1997) Use of nitrogen simulation models for site-specific nitrogen fertilization. Precision agriculture 1997. First European Conference on Precision Agriculture, Vol. I. Spatial variability in soil and cropOxford, UK Bios Scientific Publishers Ltd pp. 361–369.
Ercoli L, Mariotti M, Masoni A, Massantini F. (1993) Relationship between nitrogen and chlorophyll content and spectral properties in maize leaves. European Journal of Agronomy 2 113–117.
Guérif M, Baret F, Moulin S, Bégué A. (2001) Télédétection, hétérogénéité parcellaire et gestion spatialisée des interventions techniques. In Malézieux E, Trébuil G, Jaeger M (Eds.). Modélisation des agrosystèmes et aide à la décisionMontpellier Repères pp. 303–326.
Guérif M and Duke C. (1998) Calibration of SUCROS emergence and early growth module for sugarbeet through optical remote sensing data assimilation. European Journal of Agronomy 9 127–136.[CrossRef]
Guérif M, Houles V, Makovski D, Lauvernet C. (2006) Data assimilation and parameter estimation for precision agriculture using the crop model STICS. In Wallach D, Makovski D, Jones JW (Eds.). Working with dynamic crop models. Evaluation, analysis, parameterization and applicationsAmsterdam Elsevier pp. 391–398.
Hayes JC and Privette I. (1998) Potential benefits of precision farming. Seventh International Conference on Computers in Agriculture 70–78.
Houlès V. (2004) Mise au point d'un outil de modulation intra-parcellaire de la fertilisation azotée du blé d'hiver basé sur la télédétection et un modèle de cultureSaint-Mandé (France) Université de Marne la Vallée.
Houlès V, Guérif M, Mary B, Gate P, Machet JM, Moulin S. (2006) Elaboration d'un indicateur de nutrition azotée du blé basé sur l'indice foliaire et la teneur en chlorophylle pour la préconisation de doses d'azote. In Guéréf M and King D (Eds.). Hétérogéneité parcellaire et gestion des cultures; vers une agricolture de précision. Editions QUAE pp. 179–198.
Houlès V, Mary B, Guérif M, Makovski D, Justes E. (2004) Evaluation of the ability of crop model STICS to recommend nitrogen fertilisation rates according to agro-environmental criteria. Agronomie 24 339–349.[CrossRef]
Jones JW, Hoogebroom G, Porter CH, Boote KJ, Batchelor WD, Hunt LA, Wilkens PW, Singh U, Gijsman AJ, Ritchie JT. (2003) The DSSAT cropping system model. European Journal of Agronomy 18 235–265.[CrossRef]
Jongschaap REE. (2006) Run-time calibration of simulation models by integrating remote sensing estimates of leaf area index and canopy nitrogen. European Journal of Agronomy 24 316.
Justes E, Mary B, Meynard JM, Machet JM, Thelier-Huches L. (1994) Determination of a critical nitrogen dilution curve for winter wheat. Annals of Botanny 74 297–407.
Kropff MJ, Bouma J, Jones JW. (2001) Systems approaches for the design of sustainable agro-ecosystemps. Agricultural Systems 70 369–393.[CrossRef]
Launay M and Guérif M. (2003) Ability of a model to predict crop production variability at the regional scale: evaluation for sugar beet. Agronomie 23 135–146.[CrossRef]
Lemaire G and Gastal F. (1997) N uptake and distribution in plant canopies. In Lemaire G (Ed.). Diagnosis of the nitrogen status in cropsBerlin, Heidelberg Springer-Verlag.
Lemaire G, Gastal F, Salette J. (1989) Analysis of the effect of N nutrition on dry matter yield of a sward by reference to potential yield and optimum N content. Congrès International des Herbages, Vol. 1 179–180.
Makowski D, Wallach D, Tremblay M. (2002) Using Bayesian approach to parameter estimation; comparison of the GLUE and MCMC methods. Agronomie 22 191–203.
Matthews R and Blackmore S. (1997) Using crop simulation models to determine optimum management practices in precision agriculture. Precision agriculture 1997Oxford, UK Bios Scientific Publishers Ltd pp. 413–420.
Moore GA and Tyndale-Biscoe JP. (1999) Estimation of the importance of spatially variable nitrogen application and soil moisture holding capacity to wheat production. Precision Agriculture 1 27–38.
Moran S, Inoue Y, Barnes EM. (1997) Opportunities and limitations for image-based remote sensing in precision crop management. Remote Sensing of Enviroment 61 319–346.[CrossRef]
Moulin S, Zurita Milla R, Guérif M, Baret F. (2003) Characterizing the spatial and temporal variability of biophysical variables of a wheat crop using hyper-spectral measurements, IGARSS 2003, ProceedingsToulouse, France IEEE International Vol. 4 pp. 2206–2208.
Nouvellon Y, Moran SM, Lo Seen D, et al. (2001) Combining a grassland ecosystem model with Landsat TM imagery for a ten-year simulation of carbon and water budget. Remote Sensing of Environment 8 131–149.[Medline]
Olioso A, Braud I, Chanzy A, et al. (2002) Monitoring energy and mass transfers during the Alpilles-ReSeDA experiment. Agronomie 22 597–610.[CrossRef]
Paloscia S, Macelloni G, Pampaloni P, Sigismondi S. (1999) The potential of C- and L- band SAR in estimating vegetation biomass: the ERS-1 and JERS-1 experiments. IEEE Transaction on Geoscience and Remote Sensing 8 2107–2110.
Penuelas J, Gamon J, Freeden A, Merino J, Field C. (1994) Reflectance indices associated with physiological changes in nitrogen and water limited sunflower leaves. Remote Sensing of Environment 48 135–146.[CrossRef][Web of Science]
Pierce FJ and Nowak P. (1999) Aspects of precision agriculture. Advances in Agronomy 67 1–85.
Prévot L, Chauki H, Troufleau D, Weiss M, Baret F. (2003) Assimilating optical and radar data into the STICS model for wheat crops. Agronomie 23 297–303.[CrossRef]
Ruget F, Brisson N, Delécolle R, Faivre R. (2002) Sensitivity analysis of a crop simulation model, STICS, in order to choose the main parameters to be estimated. Agronomie 22 133–158.[CrossRef]
Ruimy A, Saugier B, Dedieu G. (1994) Methodology for the estimation of terrestrial net primary production from remotely sensed data. Journal of Geophysical Research 99 5263–5283.[CrossRef]
Tarentola A. (1987) Inverse problem theory methods for data fitting and model parameter estimation Elsevier Science Publisher BV.
Ustin SL, Jacquemoud S, Zarco-Tejada PJ, Asner GP. (2004) Remote sensing of the environment: state of the science and new directions. In Ustin SL (Ed.). Manual of remote sensing, Vol. 4. Remote sensing for natural resource management and environmental monitoring John Wiley & Sons.
Verhoef W. (1984) Light scattering by leaf layers with application to canopy reflectance modeling: the SAIL model. Remote Sensing of Environment 16 125–141.
Vouillot MO, Huet P, Boissard P. (1998) Early detection of N deficiency in wheat crop using physiological and radiometric methods. Agronomie 18 117–130.
Weiss M, Baret F, Myneni R, Pragnère A, Knyazikhin Y. (2000) Investigation of a model inversion technique for the estimation of crop charcteristics from spectral and directional reflectance data. Agronomie 20 3–22.[CrossRef]
Wigneron J-P, Pardé M, Waldteufel P, Chanzy A, Kerr Y. (2004) Characterizing the dependence of vegetation model parameters on crop structure, incidence angle, and polarization at L-band. IEEE Transactions on Geoscience and Remote Sensing 42 416–425.[CrossRef]
CiteULike 
Connotea 
Del.icio.us What's this?
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||