Remotely Sensed Phenology of Coffee and Its Relationship to Yield

Transcription

1 Remotely Sensed Phenology of Coffee and Its Relationship to Yield N. A. Brunsell 1 Department of Geography, University of Kansas Lawrence, Kansa P. P. B. Pontes Department of Agricultural Engineering, State University of Campinas Campinas, São Paulo, Brasil R. A. C. Lamparelli Meteorological and Climatological Research Center Applied to Agriculture State University of Campinas, Campinas, São Paulo, Brasil Abstract: Due to complex microclimatic interactions, a biannual phenological cycle, and the generally small scale of coffee plantations, there have been few applications of satellite observations to examine coffee yield. Using data, surface precipitation and air temperature are related to MODIS surface temperature and fractional vegetation. Using lagged correlation analysis and deviations from the annual cycle, yield is related to accumulated deviations in fractional vegetation. Results imply that the coarse spatial resolution of MODIS data is compensated for by high temporal coverage, which allows for determination of coffee phenology. INTRODUCTION Coffee is one of the major agricultural products in the world, comprising, in particular, one of the major international products for several African and Latin American countries. Given the economic and societal importance of agricultural production, it is imperative to relate variations in yield to microclimatic variations in conjunction with agricultural practices. Currently, there has been some success modeling humanenvironment interactions for some of the major agricultural crops (Kaufmann and Snell, 1997; Semenov and Porter, 1995). In addition to understanding the current factors governing agricultural yield and management strategies, it is also imperative that we understand how these dynamics are likely to vary under global climate change scenarios. 1 Corresponding author; brunsell@ku.edu 1 GIScience & Remote Sensing, 2009, 46, No. 3, p DOI: / Copyright 2009 by Bellwether Publishing, Ltd. All rights reserved.

2 2 BRUNSELL ET AL. While the scientific community is moving in that direction for tropical agricultural production (Challinor et al., 2005), we are not yet there in terms of understanding the potentials of climate change on coffee production. There are several issues that must be investigated before we can understand the relationships between agricultural practices, coffee yield, and global climate change. Coffee represents a unique set of problems in that it follows a biannual phenological cycle, is often grown at small spatial scales, and has a complex relationship between microclimatic variability and final yield. These issues make it difficult to monitor with satellite platforms and utilize traditional crop yield models to assess the impacts of global change. Routine prediction of coffee yield is complicated by the fact that coffee has a biannual phenological cycle with specific microclimatic requirements at different phases of the growth cycle (Table 1). There are six phases of the biannual growing cycle of coffee (Camargo and Camargo, 2001). The first two phases correspond to vegetation production, with phase 1 (lasting from September to March) consisting of relatively long days and formation of foliar buds. The second phase lasting five months from April to August of year 1 consists of maturation and dormancy of the buds, corresponding to reduced day length and the associated decrease in air temperature (Barros et al., 1997). The second year consists of four phases in the growth cycle. The third stage is marked by flowering of the buds and lasts four months from September to December. The fourth phase is a growth phase for the fruit, lasting three months. After the fruit has reached full size, the fifth stage is a period of three months in which the fruit ripens. The sixth and final stage is senescence, which occurs in July and August of the second year. Each phase of the coffee cycle is marked by specific water requirements. The water requirements for Coffea arabica are complex (Carr, 2001), most importantly during the end of the second phase when a dry period needs to be experienced by the plants followed by a wet period to induce flowering of the buds in the third phase. This dry period can either be induced by a soil moisture limitation or through an increase in the saturation deficit at the leaf level (Carr, 2001). The saturation deficit is defined as the difference between the saturation vapor pressure at the leaf and actual vapor pressure of the air. The flowering in phase three begins approximately 8 15 days after a wet period. This wet period can be induced through either precipitation or irrigation (DaMatta, 2004a). These water relations are currently unclear, having resulted in little commercially useful relationships (Carr, 2001), possibly due to the compounding issues of evaporative demand and a lack of quantitative studies on the relationship between coffee production and saturation deficit (DaMatta, 2004a). A potentially complicating factor when considering the water requirements, and the ramifications of climate change, is the variability of stomatal conductance to net assimilation in coffee. There is evidence that this is a result of stomatal closure at high levels of irradiance, regardless of soil and leaf water status (Carr, 2001). Thus, role of management strategy of the coffee plantation becomes an important influence, particularly in the use of cropping density to control shading (Lin, 2007) and limitation of increased stomatal resistance at high irradiance levels. In general, it appears that arabica coffee transpiration and stomatal conductance is positively correlated with hydraulic conductivity (DaMatta, 2004b) regardless of plant water levels (Barros et al., 1997). In addition, Tausend et al. (2000) found that this correlation varies with the ratio of leaf area to stem area. They also found (ibid.)

3 REMOTELY SENSED PHENOLOGY OF COFFEE 3 Table 1. The Biannual Phenological Cycle of Coffee First phenological year Second phenological year First phase Second phase Third phase Fourth phase Fifth phase Sixth phase Vegetative and vegetative bud formation Long days 7 months Water restriction affects buds and following-year production Induction, growth, and dormancy of floral buds Short days evapotranspiration = 350 mm Flowering; pin-heads and fruits expand Small leaves Water restriction light bean Formation of full fruit Evapotranspiration = 700 mm Water restriction spell bean size Maturation of fruit Rest and branch senescence Water restriction good quality Sep Oct Nov Dec Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec Jan Feb Mar Apr May Jun Jul Aug Vegetative stage Rest Reproductive stages Pruning Source: Adapted from Camargo and Camargo, 2001.

4 4 BRUNSELL ET AL. that the extent of coupling between the leaf and the atmosphere (using the Ω factor of Jarvis and McNaughton, 1986) is strongly related to the stomatal conductance. The decoupling factor was shown to vary with crown architecture and therefore was interpreted to be determined primarily by boundary layer conductance at the leaf level. This highlights the role of planting density as a major factor in determining transpiration due to minimizing the role of wind on the saturation deficit. Given the economic importance of coffee production, it seems reasonable to want to use remotely sensed data to monitor coffee plantations as well as determine robust relationships for yield prediction. Remote sensing data have been routinely used in agriculture to map areas, improve yield prediction, measure production, and analyze quality. Brazil is the world s largest coffee producer and exporter. Coffee production area in Brazil totals 2.30 million hectares. Coffee is produced in 11 states, with the state of Minas Gerais, with 1.04 million hectares, accounting for 48.8% of the country s production, 99.9% of which is Arabica coffee (CONAB, 2008). However, remote sensing studies devoted to coffee in Brazil have been limited. This is due to a set of specific challenges, which ultimately involves a scale problem when using remote sensing data to monitor coffee yield. Because this crop is normally planted in small fields with considerable slope, high-resolution imagery is required to identify coffee production areas. The use of airborne imagery to develop a ripeness index for monitoring coffee has been successful (Johnson et al., 2004), as has a neural network based approach (Furfaro et al., 2007). In addition to the high spatial resolution, the biannual growing cycle (Table 1) requires high temporal coverage to accurately monitor the status of the crop at different times during the phenological cycle. This necessitates the use of satellite platforms (rather than airborne instruments) in order to capture the temporal variability, but spatial resolution becomes an important issue. Using Landsat images, spectral separation between coffee crops and woodland areas was only moderately successful in Costa Rica (Cordero-Sancho and Sader, 2005), and low classification accuracy has been observed between coffee plantations and woodland areas in Colombia (Langford and Bell, 1997). In coffee plantations in Brazil, just half of the areas identified on IKONOS-II imagery were found in Landsat/ETM+ images (Ramirez et al., 2006). High-resolution scanners may provide a solution for improving accuracy in mapping coffee areas (ibid.). In addition, the use of filters improved overall classification quality significantly in SPOT 5 imagery (Shaban and Dikshit, 2002). The monitoring of large areas under agricultural land uses and the analysis of temporal changes has been improved with the advent of Moderate Resolution Imaging Spectroradiometer (MODIS) imagery. Pax-Lenney and Woodcock (1997) have shown the necessity of using coarse-spatial-resolution imagery for the purpose of providing sufficiently frequent temporal coverage. However, it remains to seen to what extent this is feasible in terms of monitoring yield. Therefore the primary goal of this paper is to assess the feasibility of using coarse-resolution MODIS data (in particular the vegetation and surface temperature data) to monitor coffee productivity. More specifically, is the coarse spatial resolution compensated for by the high temporal coverage? In addition to this fundamental question, we are interested in determining the impact of climatic variability (in particular precipitation and temperature variability) on coffee yield. This is particularly relevant

5 REMOTELY SENSED PHENOLOGY OF COFFEE 5 Fig. 1. Monte Santo de Minas, Minas Gerais, Brazil. SPOT 5, RGB453 composite. with respect to assessing the potential impacts of regional climate change on this economically significant crop. STUDY AREA [Author: mean annual air temperature or mean monthy air temperature ] The study area is the municipally Monte Santo de Minas, in southern Minas Gerais, Brazil (Fig. 1). Based on data from the meteorological station in Monte Santo de Minas, mean annual air temperature varies from 20 to 24 C, and annual precipitation is between 1125 and 1725 mm. More than 77% of the coffee fields are 900 m or higher in elevation, offering excellent conditions for Arabica coffee production. More than 97% of coffee production is from fields with less than 15% slope, and therefore it is possible to use mechanized harvesting in these areas. To obtain elevation, slope, and aspect, we used an elevation model (90 m spatial resolution) derived from Shutter Radar Topography Mission (SRTM) data. Climate data from the COOXUPÉ Meteorological Station (consisting of air temperature and precipitation data) were also utilized. These data were compiled at

6 6 BRUNSELL ET AL. eight-day intervals to match the temporal resolution of the MODIS imagery. Annual yield data were obtained from the Cooxupé database. Mapping Coffee Fields METHODS In order to assess the ability of MODIS to monitor the coffee producing areas, for classification purposes we used an image acquired in August 2005 from the Systeme Probatoire pour l Observation de la Terre (SPOT) satellite, consisting of four multispectral bands (10 m spatial resolution) and a panchromatic band (2.5 m spatial resolution). The image encompasses the entire study area and was processed in four steps. First, the multispectral bands were merged with the panchromatic using the Gram- Schmidt Spectral Sharpening algorithm, and one archive.tif file was generated with all bands at 2.5 m spatial resolution. In the second step, the image was orthorectified using the 90 m digital elevation model derived from the SRTM. The third step consisted of registration using GPS control points collected in the municipality. Difficulties occurred in separating coffee fields and woodlands, and delimiting small fields as well as fields with young coffee. Therefore a binary classification (coffee/not coffee) was conducted. In terms of the Kappa Index of Agreement, made with 100 points sorted randomly, the mapping accuracy was excellent (0.979). The RMS error for the image after corrections was 0.78 pixels. Finally, we applied a high-pass filter (kernel size 3 3) in each band and summed the image result with each corresponding initial band without the filter. This fusion was made to enhance image texture and provided additional spectral differences between coffee fields and woodland. Other research with SPOT images has shown that adding one texture band to spectral bands improves the overall classification accuracy significantly for all datasets (Shaban and Dikshit, 2002). Classification accuracy was confirmed by visiting the coffee plantations. Data Processing To analyze temporal changes and relationships with the coffee growth cycle, we used Normalized Difference Vegetation Index (NDVI) images (MOD13Q1; maximum composite at 16 days) and surface temperature images (MOD11A2; eight-day composites) from MODIS. All images were reprojected to UTM WGS84 23 South. NDVI is a good biomass estimator and is largely used to analyze agricultural changes (Jackson et al., 1986), and is based on the difference in reflectance between the near infrared and red bands. Here, we convert the NDVI data into Fractional Vegetation (Fr) by: NDVI NDVI 0 Fr = NDVI MAX NDVI 0 (1) where NDVI0 is the bare soil NDVI value, and NDVIMAX is the NDVI value corresponding to maximum vegetation cover (Gillies and Carlson, 1995). Converting to Fr

7 REMOTELY SENSED PHENOLOGY OF COFFEE 7 is more beneficial than using the NDVI value for several reasons: (1) Fr represents a physically meaningful, conserved quantity that can be related to production values; and (2) Fr minimizes the need for atmospheric corrections. This second feature is particularly significant when dealing with a time series of composite temporal data in a region, where daily radiosonde profiles are not available and each pixel value may be derived from different days (e.g., would require conducting daily atmospheric corrections for the entire six-year study period). The significance of Fr being a conserved quantity is an important one when one wants to examine the joint variability between Fr and radiometric surface temperature (T) because T is collected at a resolution of 1 km and Fr is at a resolution of 250 m. Because Fr is a conserved field, spatial aggregation can be achieved by simply taking the average over neighboring pixels. This is important for our purposes inasmuch as we want to examine the general behavior of coffee fields as a function of time. Therefore, we have averaged the Fr to 1 km for comparison with the surface temperature data. One of the main objectives is to analyze and compare changes in the satellite data with the biannual coffee cycle and production. For this purpose we used Fr, precipitation observed at the weather station (PPT), and surface radiometric temperature (T), as well as the same variables with the seasonal cycle removed (Fr', PPT', and T'). To examine how temporal variability within a growing cycle impacts total production, the seasonally detrended anomaly values are accumulated within the period. The seasonal trend is quantified by averaging the six years of MODIS data for each composite period. In order to examine interannual variability within the coffee fields, we analyzed the distribution using lagged correlations between the climate data and Fr, as well as the detrended climate data and Fr. The lagged cross correlation between two time series (x,y) is calculated as: ρλ ( ) Σ[ xi () x] [ yi ( λ) y] = Σ[ xi () x] 2 Σ[ yi ( λ) y] 2 (2) where ρ is the cross correlation as a function of lag (λ) and the overbar indicates the temporal mean. The lagged correlations were calculated at a 1 km scale between the time series of Fr and T and then averaged as a function of the time lag. RESULTS AND DISCUSSION Figure 2 shows the total precipitation for each phase (e.g., see Table 1) of coffee production for the three biannual cycles considered: September 2000 to August 2002, September 2002 to August 2004, and September 2004 to August 2006 ( , , and , respectively). Precipitation is the maximum from September to March, which is Phase 1 and Phases 3 4. Outside of the rainy season, there is drastically reduced precipitation for each of the three cycles. In addition, mean air temperature is also shown in Figure 2. Temperatures are generally warmest at the beginning of the rainy season and this

8 8 BRUNSELL ET AL. Fig. 2. Air temperature and accumulated precipitation as a function of the coffee phase for each of the biannual coffee cycles considered in this study ( , , and ). results in higher temperatures in Phases 1 and 3, with temperatures decreasing from Phase 2 and 4 6. This is most pronounced in the growth cycle, which also shows higher total precipitation values. To examine if the differences between the first and second years in the growing cycle are detectable with MODIS, we examined the average surface temperature and Fr for low-production (year 1) and high-production (year 2) years over the growing season (Fig. 3). Fr is higher in the second year (Fig. 4A), most noticeably in the first half of the growing cycle (Fig. 3). This behavior can be explained using the different quantity of beans produced and the impact of harvest on the tree. In low-production years, the trees start the vegetative year getting back leaves pulled off and recovering from other injuries caused during the harvest of the previous high-production year. In the same way, during a high-production year, the injuries caused during the harvest from the prior low-production year are lower. There is less of a difference in surface temperatures, but the higher production year of the biannual period seems to correspond with slightly lower surface temperatures (Fig. 4B). Recall that a secondary objective of this paper is to determine the impacts of the timing of precipitation on the remotely sensed variables related to coffee production. Because coffee production is very sensitive to the timing of precipitation, we computed the lagged cross correlations between the seasonally detrended Fr and PPT for (Fig. 5). Lags were calculated every 16 days, with positive lags indicating that PPT leads Fr, and negative lags that Fr leads PPT. In the negative lags, there is very little correlation between Fr and PPT. The correlation generally increases as PPT begins to lead Fr, with maximum correlations occurring at 16 and 64 days.

9 REMOTELY SENSED PHENOLOGY OF COFFEE 9 Fig. 3. Averaged annual cycles in surface temperature and Fr for low- and high-production years in the coffee cycle at 16-day intervals. Radiometric temperature is often linked to soil moisture status (e.g., Brunsell, 2006), and therefore we conducted the lagged correlation analysis between Fr and T to assess the impact of soil moisture limitation and potential water stress (Fig. 5). There is a general linear trend between the cross correlation and the temporal lag (ranging from approximately 0.4 to 0.1), a similar trend to the Fr-PPT correlation. However, the correlation is always negative, which corroborates the interpretation of radiometric temperature as a proxy for soil moisture. As radiometric temperature decreases (implying more moisture) vegetation should be less water stressed and will be able to increase production. This is, of course, assuming that water is the limiting factor to growth. As Fr begins to lead the temperature field, there is a general decrease in the magnitude of the correlation. To relate the remotely sensed variability to the high and low production years associated with coffee phenology, we repeated the lagged correlation analysis for only the respective years (Fig. 6). Very little correlation exists for Fr PPT in either the high or low production years (Fig. 6A). For Fr T relationships, high production years exhibit a strong linear relationship ranging from 0.4 to 0.1 over lags of 48 to 64 days (Fig. 6B). During low-production years, the Fr T correlation shows a few peak positive correlations at 48, 32, and 80 days. This has significant importance because coffee must experience some water stress in order to increase the quality of the production, particularly in the first (low) year of the biannual cycle. Because there was not a strong correlation between Fr and PPT for coffee fields on average, we separated out each year and examined the detrended Fr and PPT. The deviation in each 16-day period was calculated as the difference for that 16-day

10 10 BRUNSELL ET AL. Fig. 4. Average Fr in low- and high-production years (A); average surface temperature in lowand high-production years (B). 1:1 lines are shown for reference. period from the six-year average. This deviation is accumulated over the growing season to examine the impacts of strong deviations from the annual average at specific times during the growth cycle compared to total accumulated Fr. The high-production years (2002, 2004, and 2006) are shown in Figure 7A. Each of these three years experienced relatively normal precipitation through the year, so the PPT deviations are small. The Fr deviations are generally high, with 2006 being the highest, followed by The lowest Fr deviation was seen in 2004, with much of the year being below average in Fr. Figure 7B illustrates the same information for the low-production years. In almost all years, the Fr are below average, which is not necessarily surprising given that these are low-production years. But the range of precipitation deviation is surprising, ranging from very dry (2001) to very wet (2003). It is interesting to note that

11 REMOTELY SENSED PHENOLOGY OF COFFEE 11 Fig. 5. Lagged correlation between detrended Fractional Vegetation (Fr) and detrended precipitation (PPT), and detrended Fr and detrended surface temperature (T). Base lags are 16 days due to the temporal compositing of the Fr data. for 2003, the additional precipitation throughout the growing season still resulted in a significantly below average vegetation response. Fr deviations are much stronger functions of temperature deviation (Figs. 8A and 8B). In 2006, the surface temperature deviation is strongly negative, which results in increased vegetation growth. In 2004, temperatures were higher than normal, which did negatively impact vegetation production. This implies that water limitation indeed does impact production, and that radiometric temperature might be a better indicator of water limitation dynamics than precipitation data. Low-production years still exhibit a relationship between temperature deviation and accumulated vegetation (Fig. 8B). The years 2001 and 2003 both exhibited higher than normal temperature deviations and resulted in lower vegetation amounts. An alternative explanation for the low-production years is that with lower vegetation, more bare soil is apparent at the 1 km scale, which will lead to a higher than normal radiometric temperature. This explanation does not explain the behavior during high-production years when there is more canopy closure. Therefore, we expect that we are seeing the impact of soil moisture in high-production years, but a combined impact of soil moisture stress and less canopy coverage in low-production years. Several things become apparent by examining the relationship between the deviations in Fr accumulated over the growing season to production values (Fig. 9). First, the biannual cycle of production is observed in the seasonally accumulated Fr. The highest yields occurred in 2002, followed by 2006 and Accumulated Fr is almost the same in 2002 and 2004, before reaching its highest value in In 2002,

12 12 BRUNSELL ET AL. Fig. 6. Averaged lagged correlation between (A) Fr and PPT and (B) Fr and surface temperature for high-production and low-production years in the phenological cycle of coffee. significant precipitation deviation was experienced in the first few 16-day composite periods, and this likely negatively impacted the accumulated Fr for that year, but not the yield, which is a result of the two-year cycle. For the three low-production years, accumulated Fr and productivity are lowest in 2003 and highest in Other than the anomalously high productivity in 2002, there is a strong relationship between accumulated deviation in Fr and yield. This implies that there is a potentially strong predictive relationship between the remotely sensed values and final coffee yield. CONCLUSIONS Monitoring the phenological development of coffee and predicting coffee yield is inherently difficult due to the biannual growing cycle. Traditional microclimatological

13 REMOTELY SENSED PHENOLOGY OF COFFEE 13 Fig. 7. Percent deviation in Fr as a function of rainfall deviation for (A) high- and (B) lowproduction periods in the coffee cycle. analysis often fails because of the complicated water and temperature requirements of the plant. Using remotely sensed data to monitor coffee in a routine manner is also complicated due to the small field size, which necessitates the use of high-resolution imagery. Typically the use of high-resolution imagery (e.g. Landsat) limits the temporal resolution that is necessary for monitoring the phenological development of coffee. Here, we have assessed to what extent the higher temporal resolution and coarser spatial resolution of MODIS satellite data can be used to monitor and predict coffee yield. Even when aggregated to 1 km resolution, MODIS data appears capable of successfully monitoring coffee. There are, of course, limitations. Six years of data encompass only three growing cycles, which limits the extent to which generalities can be made. In addition, given only one site for precipitation measurement in the area, we are forced to assume that the observed precipitation at that site is valid for

14 14 BRUNSELL ET AL. Fig. 8. Average lagged correlation between deviation in Fr and deviation in surface temperature for high-production (A) and low-production (B) years. the entire area. However, by focusing on the correlation structure, we are not so much interested in the actual value of the precipitation, but rather the deviations. By focusing on the deviation from annual trends in the remotely sensed fields, we have shown that accumulated fractional vegetation generally reflects the trends in coffee yield. When combined with deviations in precipitation and surface temperature, we can explain how water limitations occur at specific times and thus augment a predictive approach to coffee yield. Therefore, we can conclude that coarse-resolution satellite data from sensors such as MODIS are capable of monitoring coffee yield. In addition, we are able to quantify the impacts of inter-annual variability in climate on coffee production.

15 REMOTELY SENSED PHENOLOGY OF COFFEE 15 Fig. 9. Annual coffee production (in kg/ha) and deviation from annual mean accumulated Fr for each year. ACKNOWLEDGMENTS Funding for this research was provided by Cooxupé Ltda. (the Guaxupé Regional Cooperative of Coffee Producers, Minas Gerais, Brazil) and the Department of Geography and College of Liberal Arts and Sciences at the University of Kansas. REFERENCES Barros, R. S., Mota, J. W. S., DaMatta, F. M., and M. Maestri, 1997, Decline of Vegetative Growth in Coffea arabica L. in Relation to Leaf Temperature, Water Potential and Stomatal Conductance, Field Crops Research, 54: Brunsell, N. A., 2006, Characterization of Land-Surface Precipitation Feedback Regimes with Remote Sensing, Remote Sensing of Environment, 100: Camargo, Â. P. D. and M. B. P. D. Camargo, 2001, Definição e esquematização das fases fenológicas do cafeeiro arábica nas condições tropicais do Brasil, Bragantia, Campinas, 60: Carr, M. K. V., 2001, The Water Relations and Irrigation Requirements of Coffee, Experimental Agriculture, 37:1 36. Challinor, A. J., Wheeler, T. R., Slingo, J. M., and D. Hemming, 2005, Quantification of Physical and Biological Uncertainty in the Simulation of the Yield of a Tropical Crop Using Present-Day and Doubled CO 2 Climates, Philosophical Transactions of the Royal Society B, 360:

16 16 BRUNSELL ET AL. CONAB (Compania Nacional de Abastecimento), 2008, Acompanhamento da Safra Brasileira, [ Cordero-Sancho, S. and S. A. Sader, 2005, Spectral Analyses and Classification Accuracy of Coffee Crops using Landsat and a Topographic-Environmental Model, International Journal of Remote Sensing, 28: DaMatta, F. M., 2004a, Ecophysiological Constraints on the Production of Shaded and Unshaded Coffee: A Review, Field Crops Research, 86: DaMatta, F.M., 2004b, Exploring Drought Tolerance in Coffee: A Physiological Approach with Some Insights for Plant Breeding, Brazilian Journal of Plant Physiology, 16:1 6. Furfaro, R., Ganapol, B. D., Johnson, L. F., and S. R. Herwitz, 2007, Neural Network Algorithm for Coffee Ripeness Evaluation Using Airborne Images, Applied Engineering in Agriculture, 23: Gillies, R. R. and T. N. Carlson, 1995, Thermal Remote Sensing of Surface Soil Water Content with Partial Vegetation Cover for Incorporation into Climate Models, Journal of Applied Meteorology, 34: Jackson, R. D., Pinter, P. J., Reginato, R. J. and S. B. Idso, 1986, Detection and Evaluation of Plant Stresses for Crop Management Decisions, IEEE Transactions on Geosciences and Remote Sensing, 24: Jarvis, P. G. and K. G. McNaughton, 1986, Stomatal Control of Transpiration: Scaling up from Leaf to Region, Advances in Ecological Research, 15:1 49. Johnson, L. F., Herwitz, S. R., Lobitz, B. M., and S. E. Dunagan, 2004, Feasibility of Monitoring Coffee field Ripeness with Airborne Multispectral Imagery, Applied Engineering in Agriculture, 20: Kaufmann, R. K. and S. E. Snell, 1997, A Biophysical Model of Corn Yield: Integrating Climatic and Social Determinants, American Journal of Agricultural Economics, 79: Langford, M. and W. Bell, 1997, Land Cover Mapping in a Tropical Hillsides Environment: A Case Study in the Cauca Region of Colombia, International Journal of Remote Sensing, 18: Lin, B. B., 2007, Agroforestry Management as an Adaptive Strategy Against Potential Microclimate Extremes in Coffee Agriculture, Agricultural and Forest Meteorology, 144: Pax-Lenney, M. and C. E. Woodcock, 1997, Monitoring Agricultural Lands in Egypt with Multitemporal Landsat TM Imagery: How Many Images Are Needed?, Remote Sensing of Environment, 59: Ramirez, G. M., Zullo Junior, J., Assad, E. D., and H. S. Pinto, 2006, Comparação de dados dos satélites Ikonos-II e Landsat/ETM+ no estudo de áreas cafeeiras, Pesquisa Agropecuaria Brasileira, 41: Semenov, M. A. and J. R. Porter, 1995, Climatic Variability and the Modelling of Crop Yields, Agricultural and Forest Meteorology, 73: Shaban, M. A. and O. Dikshit, 2002, Evaluation of the Merging of SPOT Multispectral and Panchromatic Data for Classification of an Urban Environment, International Journal of Remote Sensing, 23: Tausend, P. C., Meinzer, F. C., and G. Goldstein, 2000, Control of Transpiration in Three Coffee Cultivars: The Role of Hydraulic and Crown Architecture, Trees, 14: