An exploratory study to improve the predictive capacity of the crop growth monitoring system as applied by the European Commission

Research output: Thesisexternal PhD, WU

Abstract

<p>The European Union (EU), through its Common Agricultural Policy (CAP), attempts to regulate the common agricultural market to, among others, secure food supplies and provide consumers with food at reasonable prices. Implementation and control of these CAP regulations is executed by the Directorate General for Agriculture (DG VI) of the EU. To manage this common market, to evaluate the consequences of these regulations and to estimate and control the subsidies to be paid, DG VI requires detailed information on planted area, crop yield and production volume.</p><p>Information on land use, interannual land use changes and yields is routinely collected by the national statistical services, which convey this information to the statistical office of the European Commission, EUROSTAT. Collection and compilation of these agricultural statistics however, is time consuming and laborious; it often takes up to one or two years before this information is available in the EUROSTAT databases. At this late stage, these statistics are of limited use for evaluating policy or to estimate the amount of subsidies to be paid. Hence, more timely and accurate information is needed. To assist DG VI and EUROSTAT to collect this information, the MARS project was initiated, with the aim to develop methods to produce timely statistics on land use, planted area and production volumes for various crops within the EU.</p><p>The MARS project applies remote sensing imagery and ground surveys to estimate the planted area. Since no proven methods to relate satellite imagery to quantitative crop yields were available at the beginning of the MARS project, a crop growth monitoring system (CGMS) based on the WOFOST crop growth simulation model was developed.</p><p>In this thesis several variants of the current standard operational version of CGMS are explored. The standard CGMS version assumes that yield per unit area and planted area are independent of each other. In this thesis total production volume instead of yield per unit area is considered, hypothesizing that the annually planted area and the yield per unit area are mutually dependent and should therefore be analyzed simultaneously. It is assumed that weather and economic factors affect production volume variation. However, for two of the major wheat producing countries the analysis fails to demonstrate a relation between the soft wheat production volume and selling or intervention price. Furthermore, for soft wheat, for 5 out of the 10 investigated countries, and for durum wheat, for 3 out of the 4 investigated countries, the expenditure on crop protection agents is not significantly associated with the production volume. These results suggest that these parameters are not generally applicable and should therefore not be applied for production volume prediction. As an alternative to economic factors, the fertilizer application per unit area is examined. The analysis shows that this factor can account for the trend and production volume variation.</p><p>Next, production volumes of soft and durum wheat are predicted and two types of prediction models were examined. The first type included the planted area in the prediction model, and production volume was predicted in one step. The second type predicted the production volume in two steps: first, yield per unit area was predicted and subsequently, this value was multiplied by an estimate for the planted area. Furthermore, two functions to describe the trend in yield and production volume series were tested: a linear function of time and a linear function of the fertilizer application. A hypothetical and an operational situation were studied. The hypothetical situation assumes that current year's information on planted area and fertilizer consumption is available, whereas the operational situation assumes that these two variables are not available and consequently have to be estimated.</p><p>Comparison of the results from the one-step model with those from the two-step model demonstrates that in the operational situation in 14 out of 16 crop-country combinations the one-step model predicted more accurately when a linear time trend was applied. When fertilizer application was applied the one-step model in 10 out of 16 crop-country combinations provided more accurate results. Furthermore, when two-step prediction models were applied, crop simulation results were significant in approximately 30% of the cases (5% t-test). However, when models of the one-step type were used, this number increased to more than 80%.</p><p>Although these results cannot be viewed as a proof that one-step models are really superior, they still give an indication and provide a direction for further research. It corroborates the assumption that variation in planted area and yield per unit area are not independent and therefore variation in production volume should be analyzed using models of the one-step type.</p><p>Comparison among the one-step model results in the operational situation shows that in 50% of the investigated crop-country combinations the model that applied simulation results plus either a linear time trend or fertilizer application, predicted more accurately than the model that did not apply simulation results. In the hypothetical situation the two-step model that uses the fertilizer application provided the most accurate results. However, analysis also demonstrates that in the operational situation this model yielded the least accurate results. In this situation, the one-step models provided the most accurate results since they are less sensitive to errors in the planted area estimates.</p><p>Although the prediction results obtained with simulation results are not always more accurate when compared to results derived from trend extrapolations or simple averages, the use of simulation results in combination with a trend function certainly holds a promise for further improvement.</p><p>Next, a method to estimate daily global radiation was developed and tested. This method uses cloud cover and the temperature range as input. It provides less accurate results than the Ångström-Prescott equation, but the differences are small. This method may be used as an alternative for the Ångström-Prescott method when sunshine duration observations are not available. A hierarchical method is proposed to introduce global radiation in CGMS. If observed global radiation is available it will be used, if only sunshine duration is available the Ångström-Prescott method will be used, if neither radiation nor sunshine is available, the method developed here may be applied. This method was tested and the prediction results were slightly more accurate than the results obtained with the standard operational version of CGMS.</p><p>Furthermore, an additive and a multiplicative model are compared. An additive model assumes that variation in production volume as a result of weather variation is similar under high production systems and low production systems. The multiplicative model assumes that variation in production volume over the years is proportional to the mean production level. Wheat production volumes for France were predicted at subregional, regional and national level. The predictions at subregional and regional level were aggregated to national values.</p><p>The results suggest that more accurate predictions of total national production volume can be obtained when predictions executed at regional or subregional level are aggregated into a national value instead of estimating this value in one step. This may be the result of the applied aggregation procedure. Presumably, local weather effects are obscured in the aggregated values. Another explanation could be that errors in the production volumes of the individual regions or subregions compensate each other when summed to a total national value. These results also provide some evidence that aggregated predictions derived from the multiplicative model are more accurate than those derived from the additive model, suggesting that effects of weather on crop growth depend on the magnitude of the annual mean yield.</p><p>Finally, data obtained from the field surveys executed in the framework of the MARS are analyzed with the aim to increase insight in sowing strategies of rainfed barley in semi-arid regions. The hypothesis is that in CGMS sowing date variation should be accounted for: CGMS assumes per crop and per region one sowing and one flowering date, hypothesizing that sowing and flowering date variation have limited effects on the regional production volume. The results, at least for barley grown under rainfed conditions, support this hypothesis: no association could be demonstrated between (i) sowing date variation and yield per unit area; (ii) sowing date variation and the precipitation amount; (iii) flowering date variation and yield per unit area. Farmers may base their sowing strategy on the fact that sowing at the presumed beginning of the rainy season will give higher yields than when sowing is delayed, provided rainfall during the growing season is sufficient. In dry years, when available water is the main yield-limiting factor, effects of sowing date variation on yield are not noticeable. The need to synchronize seasonal rainfall and phenology of the selected barley cultivars may also limit the possibilities to postpone sowing.</p><p>Evaluation</p><p>The principal objective of this study was to explore possibilities to improve CGMS in such a way that it may be applied for quantitative yield prediction for all EU member states. Various options have been explored. Although some interesting results have been obtained, only two concrete suggestions for such an improvement can be given: (i) predictions should be executed at lower administrative level and subsequently aggregated to national values, (ii) planted area should be included in the analysis and prediction model. More research is needed to identify tangible points for improvements in CGMS.</p>
Original languageEnglish
QualificationDoctor of Philosophy
Awarding Institution
Supervisors/Advisors
  • van Keulen, H., Promotor
  • Jansen, M.J.W., Promotor
Award date7 Jan 2000
Place of PublicationS.l.
Print ISBNs9789058081858
Publication statusPublished - 2000

Keywords

  • wheat
  • crop yield
  • sowing date
  • solar radiation
  • yield forecasting
  • simulation models
  • geographical information systems
  • growth
  • european union

Fingerprint Dive into the research topics of 'An exploratory study to improve the predictive capacity of the crop growth monitoring system as applied by the European Commission'. Together they form a unique fingerprint.

Cite this