to be generally promising, although in a few selected cases reseachers 
have demonstrated limited success. The problem is that often the char- 
cateristic of the crop which contributes the bulk of the information re- 
corded by remote sensors, namely foliage density and condition, is not 
highly correlated with the characteristic of interest, such as grain, 
flower, or fruit production. For example, a grain field may possess lush, 
dense leaf cover, but if temperatures and available soil moisture are 
not suitable during certain critical periods, the grain itself may not 
develop in an optimum manner. The direct measurement approach has been 
more encouraging for crops where the foliage itself is the item of in- 
terest, such as in forage crops, and in assessing the effects of phys- 
ically destructive factors, such as hail damage, lodging, or flooding 
on yield reduction. 
On the other hand, agro-meterological modeling has demonstrated 
considerable potential usefulness for yield forecasting, although in 
general it is still far from operational. It is quite possible that the 
yield modeling efforts being carried out in the LACIE project may result 
in significant breakthroughs in this area. Nevertheless, the whole sub- 
ject of yield forecasting is one in which much more knowledge is needed 
before the contribution of remote sensing to crop production forecasting 
can be fully realized. 
Soil effects 
The appearance of vegetation when viewed from the aerial view is 
affected to a great extent by the underlying soil. This may be due to 
varying plant specie mixtures on various soils, variations in plant cover 
and/or vigor due to soil effects, or the color of the soil itself as it 
contributes to the radiance signature where less than 100% plant canopy 
cover exists. 
Whatever the cause, a much better understanding is needed of plant- 
soil relationships as they effect the recording of vegetation appear- 
ance by means of remote sensing. Such aspects of vegetation classifi- 
cation as training sample selection and signature extension are intimately 
tied to the variation in the scene, and their usefulness is dependent on 
a knowledge of soil effects. Hopefully the accuracy both vegetation 
classification and soils mapping would be improved given such information. 
Forest conditions 
LANDSAT imagery has proven useful in delineating forestland from 
nonforestland when imagery at the proper time of year is used for inter- 
pretation. Considerable research efforts have focused on problems assoc- 
iated with extracting more detailed information within the forestland 
class. Although it is evident that detailed forest type mapping by species 
cannot be accomplished consistently with LANDSAT data, stratification of 
forest stands into strata related to density class, condition class, age 
class, etc. may be possible with controlled clustering techniques. 
Research is needed to define the relationship between LANDSAT spectral 
signatures of forest stands and such parameters as slope, aspect, ele- 
vation, stand density in terms of crown closure and basal area, size class, 
age class, and timber volume. In most forested areas, the correlation of 
spectral signatures with parameters related to stand conditions will vary 
as a function of phenology and time of year. Definition of these correl- 
ations will aid in specifying a statistical analysis approach with clust- 
ering techniques and will provide guidance as to the level of detail which