Spectral Estimates of Solar Radiation Intercepted by Corn Canopies

As the need for timely information on worldwide crop production intensifies the role of remote sensing is becoming more prominent. If agronomic variables related to yield could be reliably estimated from multispectral satellite data, then crop growth and yield models could be implemented for large areas. The objective of this experiment was to develop methods for combining spectral and meteorological data in crop yield models that are capable of providing accurate estimates of crop condition and yields. Initial tests of this concept using data acquired in field experiments at the Purdue Agronomy Farm, West Lafayette, Ind., are presented. Reflectance factor data were acquired with a Landsat band radiometer throughout two growing seasons for corn (Zea mays L.) canopies differing in planting dates, populations, and soil types (Typic Argiaquoll and Udollic Ochraqualf). Agronomic data collected to coincide with the spectral data included leaf area index (LAI), biomass, development stage, and final grain yields. The spectral variable greenness was associated with 76% of the variation in LAI over all treatments. Single observations of LAI or greenness were found to have limited value in predicting corn yields. The proportions of solar radiation intercepted (SRI) by these canopies were estimated using either measured LAI or greenness. Both estimates, when accumulated over the growing season, accounted for approximately 65% of the variation in yields. The Energy Crop Growth (ECG) variable was used to evaluate the daily effects of solar radiation, temperature, and moisture stress on corn yields. Coefficients of determination for grain yields were 0.67 for the ECG model using measured LAI to estimate SRI, and 0.68 for the ECG model using greenness to estimate SRI. We conclude that this concept of estimating intercepted solar radiation using spectral data represents a viable approach for merging spectral and meteorological data in crop yield models. The concept appears to be extendable to large areas by using Landsat MSS data along with daily meteorological data and could form the basis for a future crop production forecasting system. Additional index words: Remote sensing, Reflectance, Grain yield, Crop models, Zea mays L. I recent years the world's food situation has emphasized the need for timely information on worldwide crop production. Remote sensing from aerospace platforms can provide information about crops and soils that could be useful to crop production forecasting systems. The feasibility of utilizing multispectral data from satellites to identify and measure crop area has been demonstrated (14), however, relatively little research has been conducted on developing the capability of using multispectral data to provide information about crop condition and yield. If this spectrally derived information can be combined effectively with crop models which depict limitations imposed on crop yields by weather and climate, then better information about crop yield and production can be gained. Solar radiation is the source of energy for photosynthesis, the initial process that green plants use to convert CO2 and water into simple sugars. Other plant processes convert these initial products of photosynthesis into dry matter (DM) including carbohydrates, proteins, and oils. Solar radiation is available as an energy source for plants only when it interacts with leaves. In a healthy crop adequately supplied with water, the production of dry matter is proportional to the solar radiation intercepted by the canopy. Thus, important components of growth and yield are the amount and duration of plant surface available for photosynthesis (2,4). In theory, the production of DM over time period t, beginning at emergence and ending at maturity, can be related to the proportion (P) of the incident solar radiation (SR) intercepted by the crop using the following 1 Contribution from the Laboratory for Applications of Remote Sensing and Dep. of Agronomy, Purdue Univ. Journal Paper No. 8978. Purdue Agric. Exp. Stn., West Lafayette, IN 47907. This study was supported by NASA Johnson Space Center Contract NAS9-15466. Received 5 Apr. 1982. 2 Research agronomist, graduate assistant, and senior research agronomist, respectively. 528 AGRONOMY JOURNAL, VOL. 75, MAY-JUNE 1983 equation from !Steven (19): DM = JmatureE emerge P SR d t . E is t h e efficiency of conversion of solar energy into DM and typically ranges from 1 to 3 g/MJ (15). This equation can be used t o predict DM production if P is known. Curren t me:thods t o measure t h e radiation intercepted by crops are laborious a n d limit use of such crop models to small resea:rch plots. If t h e proportion of energy available for crop growth could be estimated reliably using multispectral satellite da ta , then t h e capability t o es t imate crop production for large regions should be improved significantly. In practice, a l though solar radiation is essential for photosynthesis, it is only one of several factors interacting t o influence: crop yields. O t h e r factors essential to crop growth and yield are water , temperature , nutrients, and COz. A n y serious and comprehensive effort to es t imate crop yields must assess t h e impact of these other factors. O u r overall objective is to develop methods for combining spectral a n d meteorological d a t a in crop yield models which are capable of providing accura te estimates of crop condition and yields throughout the growing season. This paper presents our initial tests of these concepts using spectral a n d agronomic d a t a acquired in controlled experiments. Future research will extend those methods t h a t best es t imate yields at agricultural experiment stations t o large areas using spectral d a t a acquired by satellites. MATERIALS AND METHODS Design of Experiment Spectral and agronomic data used in these analyses were acquired at the Purdue University Agronomy Farm in 1979 and 1980. A full season corn (Zea mays L.) hybrid, Beck 65X, was grown on two soil types. When dry, the Chalmers silt loam (fineloamy, mixed, mesic Typic Argiaquoll) had a dark gray (IOYR 4/ 1) surface while the Toronto silt loam (fine-silty, mixed, mesic Udollic Ochraqualf) had a light gray (IOYR 6/1) surface. These two soils were spectrally distinct in both visible and near infrared reflectance factor (1 1). Prior to planting 200, 50, and 95 kg/ha of N , P, and K, respectively, were applied uniformly to both soils to minimize the risk that corn growth might be limited by nutrient availability. Daily meteorological data were recorded a t the cooperative National Weather Service station (West Lafayette 6NW) which was within 200 m of the fields. Incoming solar radiation was measured with an Eppley Precision Spectral Pyranometer and recorded as total M J m-2 day-’ (1MJ = 2.387 X 10-l’ Langley). Daily maximum and minimum air temperatures were measured with liquid-in-glass thermometers in a standard Cotton Region shelter. Soil moisture in the top 105 cm (depth of drainage tiles) was estimated on a daily basis using a soil moisture balance model (20). Inputs to the soil moisture balance model include initial moisture content, water holding capacity, and wilting p i n t moisture content of each soil type plus daily measurements of maximum and minimum air temperatures, precipitation, and evaporation from a standard class A pan. Within each soil type two completely randomized blocks with three plant densities (25,000, 50,000, and 75,000 plants/ha) were planted in 76 cm wide north-south rows on 2, 16, and 30 May 1979 aad 7, and 22 May, and 11 June 1980. Additional plots of the 50,000 plants/ha treatments were planted on 16 and 29 May: 18 June, and 3 July 1980. These treatments represented a wide range of planting dates and plant populations expected in t m n fields in Indiana. Canopy Characterization Agronomic variables measured weekly included: plant height, leaf area index (LAI), development stage (8), total fresh and dry biomass, stalk (including leaf sheath), ear, and green leaf blade (lamina) dry weights. Percent crop cover (defined as the percentage of soil covered by vegetation) was determined by placing a grid over a vertical photograph and counting the intersections occupied by green vegetation. Crop biomass was estimated by harvesting three plants in 1979 and four plants, in 1980 from each plot. Each sample was weighed immediately, separated into its components, dried a t 75 C, and reweighed. The area of a random subsample of green leaf blades from eiich plot was measured with an electronic area meter (LI-COR, Model LI13000) and the green leaf area to leaf dry weight ratio was calculated. Leaf area index was calculated using this leaf area/weight ratio, the total dry weight of green leaves from all plants sampled, and the soil area represented. Grain was harvested by hand from the center four rows of each plot (1 1.6 m2), dried, weighed, and corrected to 15.5% moisture. Visual assessment of the soil moisture and crop condition were made during the spectral data collection. Crop condition assessment included evaluations of lodging and hail and insect damage:. Spectral Measurements Radiance measurements, used to determine reflectance fac:tor (RF), were acquired with a Landsat-band radiometer (Exotech 100) throughout the growing season in each year. Robinson and Biehl (17) describe the conditions and procedures for obtaining the R F data. The Exotech 100 has a 15” field of view itnd acquired data in the following wavelength regions: 0.5 to 0.6, 0.6, to 0.7, 0.7 to 0.8, and 0.8 to 1.1 pm. Data were taken only when there were no clouds over or in the vicinity of the sun and when the solar elevation was at least 45” above the horizon. The radiometer and a 35-mm camera were attached to a boom mounted on a pickup truck and elevated 5.2 m above the soil in 1979 and 7.6 m in 1980. After the instruments were leveled for a nadir look angle, two measurements were takenone centered over the row and one centered between rows--in each plot to better estimate the overall canopy response (5). A color photograph which included the area viewed by the rz.diometer was taken vertically over each plot and used to determine percent crop cover. Analysis Two methods of estimating the proportion of solar radiation intercepted by corn canopies were examined. First, the proportion of intercepted radiation (SRIL) was described as a function of measured LA1 using the following equation from Linvill et al. (12): This is an application of Bouguer’s Law using LA1 and an extinction coefficient of -0.79. When LA1 is 0, no energy is intercepted. When LA1 is 2.8, approximately 90% of the visible solar radiation is intercepted by the canopy and is potentidly useful to the crop. The second method estimates SRIL as a function of the spectral variable greenness. This spectrally estimated proportion of radiation intercepted is called SRIs to distinguish it from SIUL which is estimated using measured LAI. The following equation was developed using data from both years of this study: where G is the green vegetation index or “greenness” for reflectance factor data (1 3). Greenness was calculated as follows: Greenness = (-0.4894 RFI -0.6125 RF2 + 0.1729 RF3 -I0.5953 RF4), where RF1 to RF4 refer to the reflectance fac:tor in each of four bands of the radiometer. SRIL = [ I exp(-0.79 LAI)] . PI SRIs = -0.1613 + 0.0811 G 0.0015 G2 [3] DAUGHTRY ET AL.: SOLAR RADIATION INTERCEPTED BY CORN Table 1. Regression analyses for LA1 and SRI in 1979 and 1980 Variable Estimator(s)t R* F RMSES CV(%)5 LA1 G 0.74 2270.0 0.88 48.6 G, G' 0.76 1245.9 0.85 47.0 G. G', D, P. S 0.79 605.4 0.79 43.6 SRI G 0.86 5063.6 0.14 24.6 G, GI, D, P. S 0.91 1565.3 0.11 20.3 (n = 811). G, G' 0.90 3632.1 0.12 21.0 t G = greenness, G' = (greenness)'. D = plant date, P = plant density, $ Root mean square error. 5 Coefficient of variation. S = soil type. Solar azimuth and zenith angles, spectral properties of canopy elements, leaf area index, leaf angle distribution, leaf size and shape, and leaf movement due to wind, wilting, and phototropism influence the interception of radiation by vegetation (16). The concept of a simple exponential extinction of radiation appears to be widely applicable. Norman (16) presents a summary of extinction coefficients reported for various canopies and sun elevation angles. For corn the extinction coefficients ranged from 1.5 for low sun angles (IO") to -0.56 for high solar elevation angles (70'). The extinction coefficient used in this research is within the range of values presented by Norman for approximately 45' solar elevation. Additional research is needed to characterize and model the changes in extinction coefficient if this approach is to be used quantitatively. SRIs, predicted as a function of greenness (Eq. [3]), and SRIL, predicted as a function of measured LA1 (Eq. [2]), were calculated for each day that appropriate spectral and agronomic data were acquired and linearly interpolated for intermediate days throughout the growing season for each plot. Daily values of SRIs and SRIL were accumulated from planting to physiological maturity. Direct correlations of final grain yields with these accumulated indices were examined. To account for variability in temperature and plant water status, the performances of spectrally estimated SRIs and measured SRIL were compared using the Energy Crop Growth (ECG) model (3,4) which combines the concept of intercepted solar radiation with a moisture stress term and a temperature function. The ECG model used was: mature ECGL = C (SRi/LE)(SRILi)(WFi)(FTi) [41 i-planted where S R is the daily solar radiation in MJ m-z dayL-' and LE is the approximate latent energy of water 2.5 X IO3 MJ m-3, WF is the ratio of daily evapotranspiration to potential evapotranspiration (ET/PET) (19), and FT is a daily temperature function (3). For the spectral ECGs model SRIs was calculated using Eq. [3] and substituted directly for SRIL in RESULTS AND DISCUSSION Relation of Canopy Reflectance to LA1 and SRI The LA1 and SRI for these corn canopies were described as functions of several spectral variables and transformations using regression analyses. Previous research has indicated that greenness is highly correlated to LA1 and percent crop cover but relatively insensitive to soil color (1 1,21). In each year and for the combined data, greenness predicted SRI better (higher R2) than LA1 (Table 1). The response of greenness to LA1 appears asymptotic for LA1 greater than approximately 5 (Fig. 1). This is consistent with infinite reflectance of single leaves where visible reflectance was minimized with two layers of leaves Eq. 141. X W n z a W n a LL