For decades scientists have sought to develop regionally applicable estimators of crop yield using models formulated from remote sensing data. With a few exceptions, most broad scale models, based on remote sensing, have used the Advanced Very High Resolution Radiometer (AVHRR) Normalized Difference Vegetation Index (NDVI) to derive retrospective, empirical relationships between NDVI and yield. However, while retrospective analyses provide insight into past performance, they do little to satisfy the need for near real time yield information. Application of these empirical NDVI models is limited to the regions and time frames for which the regression equations were formulated. This means that regression models must be carefully re-evaluated each season, limiting their practical utility. Unlike crops whose yield consists of above ground production, wheat yield is contained in storage organs, and is very sensitive to adverse meteorological conditions at critical growth stages, dictating that grain yield must be modeled and not inferred. To this end our study was designed to 1) assess the potential of MODIS GPP for estimating wheat yield in Montana and North Dakota and 2) define the practical limits within which wheat yield can be sufficiently estimated using these data. To achieve these objectives MODIS GPP data were integrated over different time periods within the 2001 and 2002 growing seasons and converted to wheat yield using a simple harvest index logic, across three spatial domains including counties, climate districts and states.
The study area consists of Montana and North Dakota (Figure 1). Most wheat in Montana (> 97 %) and North Dakota (> 99%) is grown under dryland conditions (i.e. without irrigation). In Montana, most agricultural lands are located in the eastern portion of the state while in North Dakota they are distributed throughout. For this study, analysis was confined to counties with greater than 12,000 ha of wheat planted in 2001 and 2002 (figure 2).
Gross primary productivity estimates from MODIS are given in kg C m2. These units are easily converted to biomass estimates because carbon comprises roughly 50% of vegetative biomass. At physiological maturity, approximately 90% of the accumulated biomass of wheat is above ground while the rest is allocated to roots. A broad review of past research indicated that, on average, across many cultivars and types of wheat (winter, spring, and durum), a harvest index of 38% can be used to estimate the amount of grain present in above ground biomass. We included this harvest index in our wheat yield formulation: EQUATION 1 where Yieldest is the yield estimate (kg ha1), a is an arbitrary growing season end point (DOY 208, 216, 225, or 233), GPPDOY is daily gross primary productivity (kg C m2), two is a conversion factor from carbon to biomass, 0.9 (90%) is the annual proportion of GPP allocated to above ground productivity, and HI is the harvest index of 38%.
Estimated wheat yields from MODIS GPP were compared with observed yield at the county, climate district, and state levels. Only state level wheat yield analysis was sufficiently accurate. Predicted wheat yields for both Montana and North Dakota were sufficiently accurate, and never deviated more than Â± 4.5% from actual yield for the duration of the study (Table 1). This research represents a preliminary attempt to fundamentally link M ODIS GPP to wheat yield in Montana and North Dakota while defining the practical limitations to this endeavor and provides the framework for a near real time wheat yield monitor.