Crop yield forecasting on the Canadian Prairies using MODIS NDVI data
Research highlights
▶ MODIS-NDVI can be used to predict crop yields on the Canadian Prairies one to two months before harvest. ▶ However, preliminary yield forecasts can be made by late June–early July. ▶ Generally, predicted yields were within ±10% of the actual observed yields. ▶ Models have to be updated as NDVI and crop yield data become available. ▶ Combining NDVI with weather data to improve model performance is the next step.
Introduction
Grain crop production plays a vital role in the economy of the Canadian Prairie Provinces (i.e., Alberta, Saskatchewan and Manitoba), with the main grain crops being wheat, canola, barley and field peas. In 2008, the area under these four crops in western Canada was about 21 million ha, of which, 10 million ha were under wheat (Statistics Canada, 2009b). A comparison of annual wheat production (Statistics Canada, 2009a, Statistics Canada, 2009b) and wheat marketing (Canadian Wheat Board, 2008) in western Canada shows that about 75 to 80% of the wheat grown on the Prairie Provinces is exported. Given the importance of these grain crops to the economy of the Canadian Prairie provinces and Canada as a whole, early crop yield forecasting is fundamental and would go a long way in helping policy makers and grain marketing agencies such as the Canadian Wheat Board (CWB) in planning for exports.
The Normalised Difference Vegetation Index (NDVI) data derived from the Advanced Very High Resolution Radiometer (AVHHR) of the National Oceanic and Atmospheric Administration (NOAA) have been used extensively to monitor crop condition and forecast yield and subsequently production in many countries including Canada. For example, Mkhabela and Mkhabela (2000) and Mkhabela et al. (2005) developed regression models using AVHHR-NDVI data to forecast cotton and maize yield, respectively, in Swaziland and concluded that the yield of both crops could be accurately predicted at least 2 months before harvest. Unganai and Kogan (1998) reported that the vegetation conditioning index (VCI) derived from AVHHR-NDVI correlated significantly (r = 0.32 to 0.95) with maize yield in Zimbabwe during the critical grain filling stage. Similarly, Lewis et al. (1998), found that AVHHR-NDVI significantly correlated (r = 0.75, p < 0.05) with maize yield in Kenya and reported that maize production forecasts could be made a month before harvest. In Spain, Vicente-Serrano et al. (2006) combined AVHHR-NDVI data and drought indices and were able to predict wheat and barley yield four months before harvest. Their predictive models explained 88% and 82% of the variation in wheat and barley yield, respectively.
On the Canadian Prairies, Bullock (1992), Hochheim and Barber (1998), Boken and Shaykewich (2002) and Wall et al. (2008) showed the usefulness and reliability of AVHHR-NDVI data for forecasting wheat yield before harvest. Bullock (1992) reported that reliable wheat yield estimates could be made by early August, which is timely enough to be useful for the CWB operations and other users. On the other hand, Basnyat et al. (2004) investigated the relationship between AVHHR-NDVI and canola, field peas, spring wheat and durum wheat grain yield and concluded that NDVI data acquired during the period 10 to 30 July were the best for forecasting grain yield of spring-seeded crops on the Canadian Prairie. Similarly, Holzapfel et al. (2009) found that NDVI data acquired between the six-leaf stage and the beginning of flowering using a hand-held optical sensor were highly correlated to canola seed yield (R2 = 0.35; p < 0.001). The authors reported that the correlations improved to R2 = 0.36 to 0.43 when the experimental locations were categorised by soil zones; a further improvement (R2 = 0.53 to 0.67) was realised when the NDVI was divided by growing degree days (GDD) with a base temperature of 5 °C. A comprehensive list of studies that have looked at the relationship between AVHHR-NDVI data and yield for different crops can be found in Funk and Budde (2009).
Recently, studies have been conducted to relate NDVI data derived from the new Moderate Resolution Imaging Spectroradiometer (MODIS) and crop yield (Doraiswamy et al., 2004, Doraiswamy et al., 2005, Funk and Budde, 2009, Becker-Reshef et al., 2010) and also for vegetation drought monitoring (Guo and Richard, 2004, Wan et al., 2004, Gu et al., 2007, Gu et al., 2008). The advantage with MODIS is that it has a better spatial resolution (250 m) and a better radiometric calibration than AVHRR allowing more accurate crop yield forecasts (Doraiswamy et al., 2004, Doraiswamy et al., 2005, Schut et al., 2009).
Although several studies have been conducted to establish the relationship between AVHHR-NDVI and crop yield on the Canadian Prairies, no such studies have been conducted to relate MODIS-NDVI data to crop yield. Moreover, most of the studies that have related AVHHR-NDVI to crop yield on the Canadian Prairies concentrated on wheat, probably due to the fact that wheat is the largest crop on the Prairies. The objectives of this study therefore, were to: (i) evaluate the potential of using MODIS-NDVI data to forecast crop (wheat, canola, barley and field peas) yield on the Canadian Prairies and (ii) identify the best time for making a reliable crop yield forecast. The ultimate goal is to design a tool for agricultural drought assessment in western Canada that will include MODIS-NDVI as one of several independent variables to improve the capability to delineate the spatial extent of drought-affected areas.
Section snippets
Description of study area
The Canadian Prairies include the provinces of Alberta, Saskatchewan and Manitoba (Fig. 1) and collectively have about 30 million ha of crop land. The Prairies extend northward from 49°N to 54°N latitudes and westward from 96°W to 114°W longitudes (Boken and Shaykewich 2002) and have three distinct agro-climatic zones including sub-humid, semi-arid and arid (Fig. 2). Precipitation ranges from 300 to 500 mm per annum (Environment Canada, 2008) and is often lower than crop evapotranspiration (crop
Results and discussion
Fig. 3 shows the evolution (throughout the growing season) of the correlation coefficient (r) for the relationship between MODIS-NDVI and crop grain yields in all the agro-climatic zones. In the sub-humid zone the NDVI is highly correlated (r = 0.51 to 0.67) with grain yield (all crops) from dekad 18 through 21 (late June to July), while in the semi-arid and arid zones the highest correlation (r = 0.72 to 0.90 and 0.48 to 0.78, respectively) is from dekad 19 through 22 (early July to early August).
Conclusion
This study has shown that MODIS-NDVI can be used effectively to predict crop yields across the Canadian Prairies one to two months before harvest; however, preliminary crop yield forecasts can be made by late June in the sub-humid zone and by early July in the semi-arid and arid zones. Depending on the agro-climatic zone, the models accounted for 48 to 90%, 32 to 82%, 53 to 89% and 47 to 80% of the yield variability of barley, canola, field peas and spring wheat, respectively. In general, the
Acknowledgements
The authors express their appreciation to Dr. Klaus Hochheim, Department of Environment and Geography, University of Manitoba for assistance with mean NDVI extraction by Census Agriculture Region and to Frédéric Bédard, Statistics Canada for providing the prairie crop mask. Funding from the Canadian Foundation for Climate and Atmospheric Science through the Drought Research Initiative is gratefully acknowledged.
References (62)
- et al.
Empirical regression models using NDVI, rainfall and temperature data for the early prediction of wheat grain yields in Morocco
Int. J. Appl. Earth Obs. Geoinf.
(2008) - et al.
A generalized regression-based model for forecasting winter wheat yields in Kansas and Ukraine using MODIS data
Remote Sens. Environ.
(2010) - et al.
On the use of NDVI profiles as a tool for agricultural statistics: the case study of wheat yield estimate and forecasting in Emilio Romagna
Remote Sens. Environ.
(1993) - et al.
Crop conditions and yield simulations using Landsat and MODIS
Remote Sens. Environ.
(2004) - et al.
Application of MODIS derived parameters for regional crop yield assessment
Remote Sens. Environ.
(2005) - et al.
Phenologically-tuned MODIS NDVI-based production anomaly estimates for Zimbabwe
Remote Sens. Environ.
(2009) - et al.
Estimating spatio-temporal patterns of agricultural productivity in fragmented landscapes using AVHRR NDVI time series
Remote Sens. Environ.
(2003) - et al.
Temperature and precipitation effects on canola yields in Saskatchewan, Canada
Agr. Forest Meteorol.
(2010) - et al.
Early maize yield forecasting in the four agro-ecological regions of Swaziland using NDVI data from NOAA's-AVHRR
Agr. Forest Meteorol.
(2005) - et al.
Assessing indicators of agricultural drought impacts on spring wheat yield and quality on the Canadian Prairies
Agr. Forest Meteorol.
(2010)
A simple model of regional wheat yield based on NDVI data
Eur. J. Agron.
Crop yield estimation model for Iowa using remote sensing and surface parameters
Int. J. Appl. Earth Obs. Geoinf.
Statistical spring wheat yield forecasting for the Canadian Prairie provinces
Agr. Forest Meteorol.
Regional yield estimation for winter wheat with MODIS-NDVI data in Shandong, China
Int. J. Appl. Earth Obs. Geoinf.
Spectral responses of wheat and its relationship to agronomic variables in the Tropical region
Remote Sens. Environ.
Analysis of time-series MODIS 250 m vegetation index data for crop classification in the U.S. Central Great Plains
Remote Sens. Environ.
Large area crop mapping using time-series MODIS 250 m NDVI data: an assessment of the U.S. Central Great Plains
Remote Sens. Environ.
Drought monitoring and corn yield estimation in Southern Africa from AVHRR Data
Remote Sens. Environ.
Using satellite and field data with crop growth modelling to monitor and estimate corn yield in Mexico
Crop Sci.
Optimal time for remote sensing to relate to crop grain yield on the Canadian Prairies
Can. J. Plant Sci.
Improving an operational wheat yield model using phenological phase-based Normalised Difference Vegetation Index
Int. J. Remote Sens.
Operational estimates of Western Canadian grain production using NOAA AVHRR LAC data
Can. J. Remote Sens.
Determining the feasibility of early seeding canola in the Northern Great Plains
Agron. J.
NDVI-Crop monitoring and early yield assessment of Burkina Faso
Int. J. Remote Sens.
A five-year analysis of MODIS and NDWI for grassland drought assessment over the central Great Plains of the United States
Geophys. Res. Lett.
Evaluation of MODIS and NDWI for vegetation drought monitoring using Oklahoma Mesonet soil moisture data
Geophys. Res. Lett.
Assessing Canadian Prairie drought with satellite & climate data
Environ. Informatics Arch.
Using NOAA AVHRR data to estimate maize production in the United States Corn Belt
Int. J. Remote Sens.
Cited by (387)
Trade-offs and synergies of ecosystem services and their threshold effects in the largest tableland of the Loess Plateau
2023, Global Ecology and ConservationImproved prediction of rice yield at field and county levels by synergistic use of SAR, optical and meteorological data
2023, Agricultural and Forest MeteorologyEarly-Season forecasting of citrus block-yield using time series remote sensing and machine learning: A case study in Australian orchards
2023, International Journal of Applied Earth Observation and GeoinformationApplicability of machine learning techniques in predicting wheat yield based on remote sensing and climate data in Pakistan, South Asia
2023, European Journal of AgronomyExploring the use of Sentinel-2 datasets and environmental variables to model wheat crop yield in smallholder arid and semi-arid farming systems
2023, Science of the Total Environment