The normalised difference vegetation index obtained from agrometeorological standard radiation sensors: a comparison with ground-based multiband spectroradiometer measurements during the phenological development of an oat canopy (2025)

More than 30years ago, successful attempts were started to monitor plant growth and phenology by radiance-based vegetation indices. The idea underlying the use of vegetation indices is to characterise the spectral signature of the land-surface reflectance by one single number. Many studies have shown that the degree of land-surface greenness and other biophysical crop parameters can be described by a combination of the reflectance in the visible and the near-infrared wavelengths (e.g. Kanemasu 1974; Tucker 1979). Crops with a high chlorophyll content have a low reflectance in the visible (especially blue and red) spectral region, and a high reflectance in the near-infrared part of the spectrum. The reason for this is that visible radiation is largely absorbed by chlorophyll pigments, while near-infrared radiation is strongly scattered in plant cells. In contrast to living green vegetation, the reflectance of a bare soil and dead biomass shows a gradual increase across the spectrum without any sharp transition between the red and the near-infrared region.

Narrowband spectroradiometers are generally used for measuring the canopy reflectance with ground-based sensors. Here, as a complement to this, and as proposed by Huemmrich et al. (1999), a combination of broadband agrometeorological standard radiation sensors was applied and tested. It is shown that the phenological development of an oat crop (i.e. emergence, green-up, senescence) can be well detected by the use of complementary standard radiometer data.

The normalised difference vegetation index

During the last three decades, a large number of spectral vegetation indices to monitor the vegetation status and the growth of plants have been developed (e.g. Tucker 1979; Broge and Leblanc 2000). One of the most commonly used indices to indicate relative greenness is the normalised difference vegetation index, NDVI (Rouse et al. 1974). The NDVI is defined by a combination of two spectral reflectance components, \(\rho _{{\Delta \lambda }} \), measured in the near-infrared (NIR, 700–1,300nm) and the red (600–700nm) range of the electromagnetic spectrum according to

The actual wavelength interval, Δλ, within the NIR and red region depends on the specific bandwidths of the spectroradiometer used.

The reflectance, averaged over the time interval Δt, is given by

with \(R_{{\Delta \lambda \uparrow }} {\left( t \right)},R_{{\Delta \lambda \downarrow }} {\left( t \right)}\) the spectral flux densities of the reflected and incident radiation measured by the downward and upward viewing sensors in the wavelength interval Δλ, respectively.

Green vegetation produces a high NDVI due to the sharp contrast between ρ _NIR and ρ _red. Soils and yellowed vegetation, however, have a distinctly lower NDVI because the reflectance in the red spectral region is enhanced and that in the NIR is reduced compared to green vegetation. Consequently, the NDVI shows a response to the seasonal variation in the chlorophyll concentration (Gitelson and Merzlyak 1997) resulting in a dynamic range of 0–1.

The use of NDVI has become standard in a large number of land-surface monitoring studies. These range from the plot scale, where ground-based radiometers may be used (Jackson et al. 1983; Baret et al. 1988), to the regional (Chen and Pan 2002; Schwartz et al. 2002) and global (Justice et al. 1985; Lloyd 1990; Reed et al. 1994; White et al. 1997, besides many others) scale, both of which are covered by satellite-remote sensing.

Site and crop measurements

During the 2004 and 2005 growing seasons, radiation measurements were carried out over an oat canopy at the agricultural research station of the Deutscher Wetterdienst (DWD, German Meteorological Service) at Braunschweig (10° 26.55′ E, 52° 17.35′ N). The plot (34 ×20.5m²) was surrounded by different crops such as winter and summer wheat, rye, sugar beet and potatoes.

At weekly intervals, the total leaf-area index, LAI_tot, was measured with an LI-3000A sensor (Li-COR, Lincoln, NE). After yellow leaf segments became visible, they were separated from the leaves and the green leaf-area index, LAI_g, was measured. The LAI attributable to yellow leaves was obtained by taking the difference LAI_y=LAI_tot−LAI _g.

Additionally, twice each week, crop height data were collected and, on the basis of the BBCH-code^{Footnote 1}, phenological observations of the developmental stages were made by a well-experienced phenological observer. The BBCH code is well documented by the Biologische Bundesanstalt für Land- und Forstwirtschaft (1997).

During the whole growth period, the crop was maintained free of diseases and weeds. However, no supplementary field irrigation was carried out during prolonged spells of dry weather, so that the crop was stressed by water deficits during the summer. To detect episodes with insufficient soil water supply, the soil moisture was measured gravimetrically at weekly intervals down to a depth of 60cm, thus covering the main part of the root zone. Subsequently, these data were converted to a volumetric soil moisture quantity.

Radiation measurements

Broadband sensor

An albedometer (type CM 7B) from Kipp and Zonen (Delft, The Netherlands) was used for measuring the hemispherical-integrated incoming and reflected solar radiation (SR) in the spectral range between 305 and 2,800nm. In addition, an upward and a downward viewing quantum sensor (type LI-190 SA; Li-COR) was applied to receive the photon-flux density from the sky and from the crop in the range of photosynthetically active radiation (PAR, Δλ = 400–700nm). The number of photons was converted into energy units by 1J ≡ 4.6μmol (see also Huemmrich et al. 1999). The factory calibration coefficients were used for all sensors. Additionally, corrections for thermal offset were made. The radiometers were installed in the centre of the plot at a height of 1.1m above the canopy top (i.e. 1.1m –2.4m above ground level dependent on crop growth). The sensors ran the whole day with a data-sampling rate of 1 s and an averaging interval of 15 min.

By taking the difference between SR and PAR, one obtains the optical infrared radiation (OIR=SR−PAR; Huemmrich et al. 1999). The OIR covers the spectral regions 700–2,800nm and 305–400nm, i.e. it involves the near-infrared range (approx. 700–1,300nm), the strong water absorption bands of the middle-infrared (MIR) radiation beyond 1,300nm at 1,450, 1,950 and 2,600nm (see Guyot 1990), and a residual narrow ultraviolet region between 305 and 400nm.

For subsequent analysis, spectral flux densities (\(R_{{SR, \downarrow \uparrow }} ,R_{{PAR, \downarrow \uparrow }} ,R_{{OIR, \downarrow \uparrow }} \)) from the daylight period, with a global radiation of at least 300W m⁻², were extracted and averaged to daily mean values. This threshold was chosen arbitrarily in order to exclude low sun elevation angles that allow ρ _SR (albedo), ρ _PAR and ρ _OIR to increase during the morning and evening hours. In addition, the number of outliers associated with rain and dew events may be reduced. Water on the sensor surfaces evaporates more rapidly under conditions of \(R_{{SR, \downarrow }} > 300\,{\text{W}}\,{\text{m}}^{{ - 2}} \) compared to periods with minor radiation reception. An additional justification for this threshold is given by Duchon and Hamm (2006), who showed that, compared to a clear or partly clouded sky, an overcast sky is often associated with a lesser albedo. In their study “overcast sky” conditions are defined by a vanishing direct normal irradiance and a remaining diffuse component that reduces the downwelling shortwave radiation to, for example, \(R_{{SR, \downarrow }} \approx 200\,{\text{W}}\,{\text{m}}^{{ - 2}} \) around noon in October (see Fig.4c in Duchon and Hamm 2006).

With reference to Eq. 1, we will use ρ _OIR and ρ _PAR instead of ρ _NIR and ρ _red to calculate the broadband vegetation index, NDVI_b.

Narrowband sensor

As a reference, a narrowband multispectral radiometer (type FAL-II, non-commercially manufactured by the Institute of Technology and Biosystems Engineering, FAL Braunschweig; see Kraft et al. 2000; Behrens et al. 2004) was used for weekly point measurements made at noon. The radiometer was mounted on a 1.5m-long boom on top of a portable 2.5m mast positioned close to the standard instrumentation. Three neighbouring intra-plot sub-areas were sampled and the individual reflectance data was subsequently averaged. The measurement took about 15min. While the upward-looking sensor was equipped with a cosine receptor for measuring incoming light from the upper hemisphere, the downward-looking sensor had a limited aperture angle of 64°, resulting in a circular surveyed target area of 3m in diameter. All three sub-areas totalled approximately 21m².

The channels of the FAL-sensor are located along the green (550nm), red (670nm), NIR (780nm) and MIR (1,676nm) wavelengths of the electromagnetic spectrum. In addition, the lower (700nm) and upper (740nm) limits of the “red edge” (715nm) can be measured. The spectral ranges of the sensor channels are given in Table1. Only channels 1, 2, 5 and 6 were used for subsequent analysis.

A “green” and a “red” vegetation index (NDVI _g , NDVI _r ) was calculated using the FAL-II spectroradiometer data. The green index is based on channel 1 (550nm) and the red index on channel 2 (670nm) for ρ _red . Both indices refer to the same NIR band (channel 5, 780nm). While the red vegetation index is commonly used for vegetation-monitoring studies (e.g. Rouse et al. 1974; Broge and Leblanc 2000; Behrens et al. 2004), the green index was introduced by Gitelson et al. (1996) because, with respect to laboratory measurements at the leaf scale, it showed a higher sensitivity to chlorophyll concentration.

Crop phenology and development

Drilling of oat took place on 17 March 2004 and on 24 March 2005 (Table2). Despite the different sowing dates, mild temperatures in early spring 2005 allowed emergence (BBCH = 9) to occur nearly on the same date as in 2004 (between sowing and emergence nearly equal effective heat sums of 132.9 °d and 135.0 °d were obtained in 2004 and 2005, respectively; base temperature: 5.0 °C). At this time, the soil-moisture level was near or a little below field capacity, so that the plant stand was supplied with sufficient water (Fig.1). During the subsequent weeks in both years the crop growth proceeded similarly in time until the flowering period ended in the last half of June on DOY (day of year) 172/173. Afterwards, ripening and yellowing developed faster in 2005 than in 2004. This is probably due to physiological stress as a consequence of the low soil moisture, which dropped to the wilting point (~0.05m³ m⁻³) on DOY 174 (Fig.1).

In both years the time patterns of the LAI_g (Fig.2a) were nearly parallel during green-up until the flag leaf appeared (BBCH=37, DOY 150). However, in 2005 the LAI_g reached a slightly higher level than in the previous year. Between DOY 166 and DOY 174, in the second half of the flowering period, LAI_g and LAI_tot dropped, possibly affected by the severe soil-water deficit.

Broadband spectral measurements in the context of crop phenology

Growing season 2004

The broadband radiation measurements began 1day after sowing on DOY 78 when the ground was bare. The initial values of the OIR-, SR- and PAR-reflectance curves were roughly 0.1 (Fig.3a); the NDVI_b started with a value of 0.3 (Fig.4). Before emergence, a small peak in all three reflectance curves occurred on DOY 93. This was due to soil brightening beginning on DOY 83 after a period of rain had ended. Prolonged rain between DOY 94 and 100 rewetted and darkened the soil, leading to a minimum in the reflectance profiles at the end of the humid weather period.

Emergence occurred on DOY 95 (BBCH = 9, Table2). At this time the ground was “spotted” with very few short plant elements of 1cm length maximum. When the springtime green-up period set in (DOY 104; LAI_g=LAI_tot ≈ 0.1), ρ _OIR, ρ _SR and NDVI_b began to increase, while the PAR reflectance decreased. Senescent plant segments were first observed at the soil adjacent crop level on DOY 145 when the third node became detectable (BBCH = 33). However, the NDVI_b did not show any clear response to the start of senescence. The reason for this was that the yellow portion on the total leaf area was not greater than 3%, while the remaining 97% was physiologically active and still rising, as indicated by the increasing LAI_g. Thus, the effect of the decreasing chlorophyll concentration at the bottom of the crop on the reflectance was masked by the radiance contributions coming from the upper vital plant elements.

Between DOY 150 and 160, when the flag leaf appeared (BBCH = 37) and LAI_g peaked (LAI_g,_max = 5.7), ρ _OIR and ρ _SR attained a maximum while ρ _PAR reached a minimum. At the same time, the vegetation index hit its peak value (NDVI_b,max = 0.82). Hereafter, due to the proceeding senescence, the OIR reflectance, the albedo and the vegetation index decreased, while the breakdown of plant pigments caused a slight rise in PAR reflectance. On DOY 209, the crop status of yellow ripeness was equalled (BBCH = 87). The ρ _OIR and ρ _SR reflectance profiles reached their summer minimum while the vegetation index followed 4days later together with the maximum of ρ _PAR.

Growing season 2005

From the beginning of the measurements at the end of March until the beginning of May (DOY 90–130), when the ground was bare or sparsely covered with vegetation, the reflectance data showed considerable scatter (Fig.3b). The reason for this is the dependence of the reflectance on the soil-moisture status. The response of the OIR, SR and PAR reflectance to wet-dry weather cycles is shown in some detail for DOY 90–130 in Fig.5a. Rain events were associated with a low reflectance caused by soil darkening, and fair weather with a large reflectance induced by soil brightening. The corresponding NDVI_b was sensitive to soil-wetness effects. During soil drying between DOY 90 and DOY 95 it decreased weakly (Fig.5b) while after rainfall events it responded with a slight offset.

In 2005, with the exception of the initial part of the profiles until DOY 130, the reflectance curves are similar to those of 2004 (Fig.3b). The ρ _OIR and ρ _SR profiles peaked at BBCH = 37 (DOY 148) while ρ _PAR reached a minimum. At peak greenness, all three reflectance profiles had a magnitude comparable with that of their 2004 counterparts (Fig.3a). However, in contrast with the year before, from DOY 180 the PAR reflectance increased and the OIR reflectance decreased strongly with proceeding yellowing leading to an abrupt decline in NDVI_b (Fig.4).

As depicted by the NDVI_b time series in Fig.4, the spring green-up started on nearly the same date as in 2004 and agreed with the increase in LAI (Fig.2). In summer 2005, the NDVI_b peaked at a slightly higher level than in 2004, which is in accordance with the higher LAI_g. The early decrease in NDVI_b on DOY 180 in 2005 mirrors the premature senescence, which may be a consequence of the physiological stress induced by the severe soil-moisture situation. Note that in 2005, about 52% of the total leaf area had become yellow by DOY174, while in 2004 this percentage proportion was observed 3weeks later. Therefore, the decreasing NDVI_b profile at DOY 180 may be a response to the chlorophyll reduction, to leaf rolling and, consequently, to the increasing soil-surface signal.

Diurnal broadband reflectance and NDVI_b variation

In order to give an indication of the variability of the vegetation index and its input quantities during the course of a single day, Fig.6 depicts the 15-min profiles of NDVI_b and of OIR, SR and PAR reflectance for two fair summer days (DOY 159 = 7 June 2004, DOY 211 = 29 July 2004) between sunrise and sunset. On the first date the LAI  peaked (LAI_tot = 6.5), and the soil was fully covered by top-green biomass and a low portion (13%) of yellow plant elements near the soil surface. During the previous night and the morning hours, dew-fall was detected by capacitive wetness sensors (Hoffmann, Rauenberg, Germany), which were installed at a height of 1.0m and 2.0m above a nearby grass canopy. Their readings indicate that the upward-facing radiation sensors were covered by water droplets until 0630hours CET (CET=Central European Time). Although the reflectance profiles show roughly the expected ∪-profile shapes with a minimum near noon, strong deviations in the albedo and OIR reflectance occurred in the morning hours. These can possibly be explained by dew wetness. The NDVI_b profile is remarkably stable during most of the daylight period and displays, with the exception of shortly after sunrise, only a weak time-dependency. Figure6b shows the situation 7weeks later on 29 July 2004, when the crop was completely yellowed and the LAI_tot  (=LAI_y) had declined to 1.1 due to leaf rolling. At this time, compared to DOY 159, the NDVI_b level is distinctly lower as a response to the seasonal maximum of the PAR reflectance and the minimum of the OIR reflectance at the end of the growth period. The diurnal variation of the NDVI_b shows a slight ⋂-profile shape. In the morning hours wetting by dew was also recorded until 0600hours CET.

Due to the fact that traceable plant physiological changes elapse in a daily rather than an hourly time scale, the changing solar zenith angle and illumination conditions are responsible for variations in the diurnal ρ and NDVI time patterns (Huemmrich et al. 1999; Song 1999). In addition, water on the sensor surfaces may also affect the measurements.

Comparison between the seasonal broadband and narrowband spectral measurements

In order to check the reliability of the broadband reflectance data and of the broadband vegetation index, a comparison was made with the narrowband sensor output starting on DOY 111 in 2004 and DOY 94 in 2005. For this purpose, the individual broadband 15-min raw data set nearest to the particular time of the weekly narrowband noon readings was used. Figure7, as a counterpart to Fig.3, depicts the characteristic seasonal reflectance pattern. In both years, the oat canopy reflected less OIR than NIR radiation as long as the chlorophyll content per unit ground area was relatively high, as indicated by LAI _g  > 1.5 (DOY 120–190 in 2004, DOY 125–175 in 2005). However, when LAI_g < 1.5 and the chlorophyll concentration per unit ground area was relatively small, ρ _OIR was greater than ρ _NIR.

Throughout the main growth period of both years, as long as LAI_g > 1.5, the ρ _PAR lay between ρ _r and \( \rho _{{\text{g}}} {\left( {\rho _{{\text{r}}} < \rho _{{{\text{PAR}}}} < \rho _{{\text{g}}} } \right)} \) and NDVI_b ranged between NDVI_g and NDVI_r (NDVI_g<NDVI_b<NDVI_r) (Fig.8). At the beginning and end of the growth period, when the ground was covered by small amounts of green vegetation or by at least 50% of dead yellow vegetation (LAI_g < 1.5), the red reflectance tended to exceed the green and PAR counterparts \( {\left( {\rho _{{\text{g}}} < \rho _{{{\text{PAR}}}} < \rho _{{\text{r}}} } \right)} \), and NDVI_r tended to drop below NDVI_g and NDVI_b.

The main objective of our paper was to demonstrate that a simple configuration of standard radiation sensors commonly used at agrometeorological research stations is well suited for evaluating the broadband NDVI. In this respect our study follows that of Huemmrich et al. (1999), who introduced the method of vegetation-index related reflectance measurements via the combination of an albedometer and two PAR sensors. However, while their observations refer to a boreal forest biome, our measurements were conducted over cereals.

Multispectral reflectance data were sampled in parallel with the narrowband FAL-II sensor at weekly time intervals in order to check the results obtained with the standard instrumentation. Based on channels 1, 2 and 5, “green” and “red” vegetation indices have been calculated as a counterpart to the broadband NDVI. All three NDVI profiles show the well-known seasonal time pattern with its typical increase during the green-up phase and the decrease during senescence (Fig.8). A similar NDVI response over the whole growth period of other cereals such as wheat and barley was reported by Jackson et al. (1983) and Fischer (1994), besides many others.

In order to show the effect of averaging over broad spectral bands, the scatterplot of Fig.9 compares the NDVI_b with its narrowband analogues. As already shown in Fig.8, the NDVI_b data are restricted to a smaller range than the NDVI_r data. This agrees with the data of Huemmrich et al. (1999), who also showed that the spread in NDVI values is smaller for broadband measurements than for narrowband red measurements. Compared to NDVI_r, the NDVI_b is increased when LAI_g is small, and lowered when LAI_g is large. A linear regression provides \(NDVI_{b} = 0.339 + 0.524*NDVI_{r} \) (not plotted) with a correlation coefficient of r ² = 0.92 (standard error = 0.047). The regression coefficients are similar to those found by Huemmrich et al. (1999), whose least square fit between NDVI_b and NDVI_r is included in Fig.9 as a broken line (\(NDVI_{b} = 0.23996 + 0.69734*NDVI_{r} \)). When NDVI_b is plotted against NDVI_g a slightly better correlation is obtained (r ² = 0.94, standard error = 0.043). The regression line has a slope close to 1 and also a positive intercept (\(NDVI_{b} = 0.116 + 0.907*NDVI_{g} \), not plotted).

In order to estimate how well the broadband NDVI is reproduced by the green and red NDVIs, we use the coefficient of efficiency, c, defined by

where the bracket stands for the average of green and red index. When both growth periods are taken as a basis, c _g between NDVI_b and NDVI_g is higher (c _g = 0.82) when compared with that between NDVI_b and NDVI_r (c _r = 0.73). This suggests that NDVI_b is a suitable greenness indicator. An indication of this is given by Gitelson et al. (1996), who examined single leaves in the laboratory and found that the green NDVI was much more sensitive to the chlorophyll concentration than the red NDVI because the reflectance in the green band shows maximum sensitivity to the chlorophyll content. This is also confirmed by the model simulations of Daughtry et al. (2000).

As far as our seasonal red and near-infrared reflectance profiles are concerned, their gross shape (Fig.7) is similar to that found by a number of authors (e.g. Kanemasu 1974; Jackson et al. 1983; Baret et al. 1988; Fischer 1994; Song 1999) over different canopies such as wheat, barley and maize. With respect to the visible spectral region, and in concordance with the reflectance spectra of green vegetation (see, e.g. Ahlrichs and Bauer 1983; Gitelson et al. 1996; Guyot 1990), the reflectance in the red is lower than that in the green wavelength interval as long as the chlorophyll concentration is high, as indirectly indicated by LAI_g > 1.5 (BBCH stages 31–77). Due to its broadband characteristic, the PAR reflectance lies between the narrowband green and red reflectance \( {\left( {\rho _{{\text{r}}} \leqslant \rho _{{{\text{PAR}}}} \leqslant \rho _{{\text{g}}} } \right)} \). At the beginning and end of the vegetation cycle, when LAI_g is below 1.5, ρ _r and ρ _g are inverse to each other \( {\left( {\rho _{{\text{g}}} \leqslant \rho _{{{\text{PAR}}}} \leqslant \rho _{{\text{r}}} } \right)} \) as a consequence of the low chlorophyll content per unit ground area.

Comparing our NIR and OIR reflectance patterns (Fig.7) we find that, during the core of the vegetation cycle, ρ _OIR is below ρ _NIR. The reason is that the OIR band includes inverse contributions in the NIR and MIR spectral region. The NIR reflectance is high as a response to the large number of green leaves (high LAI_g), whereas the MIR reflectance is low because of the high leaf-water content (Guyot 1990). Consequently, broadband averaging over the NIR and MIR range shows a tendency to compensate both inverse contributions, thus leading to a smaller reflectance in the OIR than in the NIR band. Conversely, at the beginning and at end of the growing season when the ground is sparsely covered with green vegetation or fully covered with yellow vegetation, ρ _OIR is beyond ρ _NIR. When yellowing proceeds and LAI_g decreases, the NIR reflectance declines faster than OIR reflectance. The weaker decrease in OIR during senescence is a response to the increasing reflectance in the MIR region (Fig.7) as a consequence of the dehydration of senescing leaves (Gausman et al. 1976; Ahlrichs and Bauer 1983; Carter 1993).

With reference to their diurnal cycle, the broadband reflectance profiles of the oat canopy show the known solar-zenith-angle dependence, with a minimum around noon and a maximum towards sunrise and sunset when the solar-zenith angle becomes larger (Fig.6). However, the reflectance profiles are not fully symmetrical about noon and exhibit strong discrepancies shortly after sunrise. One reason for this asymmetry might be the effect of morning dew, as mentioned by Grant et al. (2000). The diurnal NDVI_b profile is remarkably smooth under fair-weather conditions.

As long as the ground is bare or sparsely vegetated, the OIR, SR and PAR reflectance is sensitive to wet-dry cycles of weather because the incoming radiation is strongly absorbed by water. Consequently, large variations in the reflectance were found in response to the changing soil brightness when the soil was alternately wetted and dried. Because the variations in OIR and PAR reflectance are similar to each other, the NDVI_b response to soil wetting and drying is weak (Fig.5). This agrees with the findings of Jackson et al. (1983), who showed that the red, green and NIR reflectance decreases when the soil becomes wet, and increases when it becomes dry. However, as is clearly shown in Fig.3b, the effect of rain on the reflectance is diminished when the ground is covered by vegetation, and only minor radiation contributions from the ground are received by the downward-looking sensors. This is confirmed by the albedo measurements of Duchon and Hamm (2006) and the spectral radiation measurements of Jackson et al. (1983).

In Fig.10, the vegetation indices are plotted against the LAI_g. The data set is split into two groups. One contains the growth subset that includes data from leaf development to heading (maximum LAI_g), and the other contains the senescence subset that includes data from heading to dead ripeness. During the growth period, the NDVIs rise with LAI_g and reach their saturation level at LAI_g = 3. Of the three NDVIs, the red NDVI shows the strongest increase from the bare-soil value towards the plateau (Fig.10a). At the end of the growth period, when senescence proceeds and LAI_g approaches zero, the NDVIs decrease. However, they remain above the initial value observed early in the growing season due to additional radiation scattering by dead foliage elements (Asrar et al. 1984). Figure10a also shows that our NDVI_r−LAI_g observations group along the two plotted NDVI_r(LAI_g) reference profiles obtained for growing and senescing wheat by Asrar et al. (1984). With respect to these reference profiles, our broadband data are enhanced in the lower LAI_g region and lowered in the upper one (see Fig.9) so that, in contrast to NDVI_r, NDVI_b is a less sensitive leaf-area indicator (Fig.10b). This is revealed in Fig.11 by the simple ratio index for the red band, \( SRI_{r} = {\rho _{{NIR}} } \mathord{\left/ {\vphantom {{\rho _{{NIR}} } {\rho _{r} }}} \right. \kern-\nulldelimiterspace} {\rho _{r} } = {{\left( {1 + NDVI_{r} } \right)}} \mathord{\left/ {\vphantom {{{\left( {1 + NDVI_{r} } \right)}} {{\left( {1 - NDVI_{r} } \right)}}}} \right. \kern-\nulldelimiterspace} {{\left( {1 - NDVI_{r} } \right)}} \), which has a stronger slope than its broadband analogue, \( SRI_{b} = {\rho _{{OIR}} } \mathord{\left/ {\vphantom {{\rho _{{OIR}} } {\rho _{{PAR}} }}} \right. \kern-\nulldelimiterspace} {\rho _{{PAR}} } = {{\left( {1 + NDVI_{b} } \right)}} \mathord{\left/ {\vphantom {{{\left( {1 + NDVI_{b} } \right)}} {{\left( {1 - NDVI_{b} } \right)}}}} \right. \kern-\nulldelimiterspace} {{\left( {1 - NDVI_{b} } \right)}} \).

Conclusions

The purpose of this paper is to show that the simple agrometeorological remote-sensing technique used here is well suited to monitoring the phenological variability of an oat crop. Two PAR sensors and an albedometer, which are combined to extract the broadband NDVI from their readings, are commonly available at agrometeorological stations and are therefore a cheap alternative compared to commercial multispectral reflectometers.

The advantage of the broadband measurements with common micrometeorological radiation instruments lies in the continuous operational mode with short time intervals, in the weather independence, and in the limited need for personnel (Huemmrich et al. 1999). In contrast, spectroradiometer measurements need more technical on-the-spot support and are therefore more time-consuming.

We have shown that the broadband NDVI profile behaves similarly in magnitude and time as its green counterpart obtained by narrowband spectroradiometer readings. In this context, one should keep in mind that the narrow and broadband NDVIs cannot be completely congruent due to the different band widths of the sensors. Model simulations of Broge and Leblanc (2000), however, show that, with respect to the canopy chlorophyll density and LAI, broadband vegetation indices are not necessarily worse predictors than narrowband indices. Nevertheless, our NDVI_b responds less sensitively to the seasonal LAI_g variation than the red NDVI. The reason is that the broadband PAR and OIR reflectance reduces the dynamic range of NDVI_b by increasing its bare-soil value and lowering its green-leaf value. As a consequence, the amplitude of the seasonal NDVI_b wave is smaller than that of the red NDVI.

The 300W m⁻² cut-off filter disregards data associated with low sun angles and is a compromise between the short-term near-noon readings (e.g. Huemmrich et al. 1999) and long-term averages taking into account nearly the complete time series between sunrise and sunset (e.g. Duchon and Hamm 2006). The restriction to sunlight periods with a solar-radiation input larger than 300W m⁻² makes it certain that water on the sensor surface may rapidly evaporate. The number of unreliable data may therefore be depleted. However, the error induced by sensor wetness was not analysed here.

The results of this study are limited to the range of surface conditions during the growth period of the oat crops. More measurements over different surface types are necessary and are currently being made over a rape crop to quantify the effect of the yellow flowering on the radiation signals. In addition, future work should address a comparison between different vegetation indices such as the NDVI, the soil-adjusted vegetation index (SAVI; Huete 1988), and others (see, e.g. Broge and Leblanc 2000) to answer the question of which of them is able to provide the most accurate information on crop phenology and vegetation properties when supplied with OIR and PAR reflectance data. For example, we found that the NDVI_b is a less sensitive indicator of LAI_g changes than the red NDVI. This clearly becomes evident when the NDVI is transformed into the simple ratio index.