R. Ra’e Ga DOI: 10.5380/raega Curitiba, v.44, p. 154 -168 , Mai/2018 eISSN: 2177-2738 SPECTRAL AGROMETEOROLOGICAL MODELING ADAPTED BY MEANS OF SIMPLIFIED TRIANGLE METHOD FOR SOYBEAN IN PARANÁ STATE – BRAZIL MODELAGEM AGROMETEOROLÓGICA ESPECTRAL ADAPTADA POR MEIO DO MÉTODO DO TRIANGULO SIMPLIFICADO PARA CULTURA DA SOJA NO ESTADO DO PARANÁ – BRASIL Daniela Fernanda da Silva Fuzzo1 ABSTRACT Agriculture is an economic activity with high dependence on weather and climate. Special geotechnology and agrometeorological modeling can be used to optimize productivity in regional and national systems, while minimizing costs. The aim was to test the agrometeorological model for estimating crop soybean yield proposed by Doorenbos and Kassam (1979), using only spectral data as input variable in the model obtained by a simplified triangle method applied in Paraná state, for crop years 2002/03 to 2011/12. A high accuracy of the data was found, the model values for the parameter d ("d " modified Willmott) were between 0.8 and 0.95, whereas the root mean 1 1 squared error showed that there was low variation between 30.81 to 116.88 (kg ha-1) and the p-value was used as the indicator significance of the model at the level of 5%, indicating that there was no statistically significant difference between the estimated and observed data, this means that the average of the data estimated by the model were statistically equal the average of the observed data. Thus, we can say that images of remote sensing can be used as tools in the absence of surface information, in agrometeorological modeling to estimate crop soybean yield. Key-words: crop yield; MODIS image; remote sensing; vegetation index. RESUMO A agricultura é uma das atividades econômica com maior dependência do tempo e do clima. Geotecnologia espacial e a modelagem agrometeorológica associadas, podem ser utilizadas na otimização da produtividade agrícola tanto em escalas regionais e nacionais, minimizando custos. O objetivo deste trabalho foi testar o modelo agrometeorológico de estimativa da soja proposto por Doorenbos e Kassam (1979), utilizando apenas dados espectrais como variável de entrada, por meio do método do triângulo simplificado. Foram analisadas regiões do Estado do Paraná e os anos-safra de 2002/03 a 2011/12. Os dados mostraram alta precisão, sendo que os valores do parâmetro d1 ( "d1" Willmott modificado) foram de 0,8 e 0,95, enquanto o erro médio quadrático mostrou que houve baixa variação entre 30,81 a 116,88 (kg ha-1 ). Em relação ao p-valor, utilizado como indicador de significância do modelo ao nível de 5%, mostrou que não houve diferença estatisticamente significativa entre os dados estimados e observados, isto significa que a média dos dados estimados pelo modelo foi igual à média dos dados observados. Dessa forma, podemos dizer que as imagens de sensoriamento remoto podem ser utilizadas como ferramentas na ausência de informações superficiais, na modelagem agrometeorológica para estimar o rendimento da soja cultivada. Palavras-chave: produtividade, imagem MODIS, sensoriamento remoto; índice de vegetação. Recebido em: 13/10/2016 Aceito em: 06/11/2017 1 Universidade Estadual Paulista “Júlio de Mesquita Filho” – UNESP, Ourinhos/SP, email: [email protected] FUZZO, D. F. S. SPECTRAL AGROMETEOROLOGICAL MODELING ADAPTED BY MEANS OF SIMPLIFIED TRIANGLE METHOD FOR SOYBEAN IN PARANÁ STATE – BRAZIL 1. INTRODUCTION Camargo et al. (1988) and Moraes et al. (1998), Agriculture is one of the most important which are agrometeorological models designed segments of the production chain and is the most to yield estimates of crop production are based dependent on natural conditions, mainly climate on climate data obtained from conventional and soil, which control the growth and weather stations. These types of development of plants. Agrometeorology aims at agrometeorological models are derived from identifying and quantifying the relationship small-scale experimental plots, which hinder the between development and yield of cultivated applicability to regional studies. Examples of plants. Once identified, meteorological elements these kinds of studies with spectral model are and the cycle period of the plants they limit, it is those of Labus et al. (2002), Mkhabela et al. possible to achieve a derivation of realistic (2005), Prasad et al. (2006), Dubreul (2005) and models for forecasting or assessing harvest Er-Raki et al. (2007). production (BERLATO et al., 1992). The Sophisticated and complex description of soybean phenology allows to mathematical models are being increasingly used identify and group the stages of development of to provide representations of the physical the crop and to relate them to their specific processes that characterize the interactions at needs during the cycle (NEPOMUCENO et al. the land surface. Attempts to describe these 2001). mechanisms and physical processes with greater Due to the importance of grains in Brazil realism led to the development of the land and internationally it is necessary to have a surface modeling systems proposed for system of crop forecasting able to identify application, for example, using images of surface production numbers in advance, as well as avoid radiant temperature (Ts) and fractional losses, and employ new technologies aimed at vegetation cover (Fr), to provide moisture reducing risk. According to Carmello et.al., (2013) content in the soil surface (Mo), and fractional among the risks, come to the question of water evapotranspiration (EF, the ratio of availability, directly interfering with the evapotranspiration ET to net radiation Rn), harvesting results, reaching the socioeconomic through a methodology known as the "triangle profile of the area, with regard to the internal method". The approach of the triangle method is and external market, reaches food security and based on a contextual interpretation of a scatter the generation of jobs and income plot derived from the relationship between Based on this, the geotechnology radiant surface temperature (Ts) and vegetation applied to agriculture has a great potential index (IV, specifically the normalized difference because through remote sensing techniques it is vegetation index NDVI) (Figure 1). possible to obtain information on estimated The discovery that the pixel envelope in acreage, crop yield and vegetative vigor, on local, Ts/NDVI (or Fr) space tends to form a triangle or state, or country scale. Thus, the objective of a trapezoid is very important since the triangular studies in remote sensing are the estimates of shape of the graph dispersion Ts / IV arises from spectral variables related to growing conditions, Ts being less sensitive to water content at the which can then be inserted into simulations of surface in vegetated areas, than in areas of productivity models (HUANG et al., 2002). exposed soil Gillies et al. (1995; 1997), Symanzik Several studies have been developed in et al. (2000), Goward et al. (2002), Sun et al. this area in order to improve and complement (2005), Arvor et al. (2007), and Brunsell and programs related to agriculture either based on Anderson (2011). These authors used different meteorological data obtained from conventional spatial datasets to demonstrate that the limits of weather stations, or even spectral models based the triangular shape (the pixel envelope depicted on satellite images. As examples, work by as the small circles in Figure 1) can be used to 155 FUZZO, D. F. S. SPECTRAL AGROMETEOROLOGICAL MODELING ADAPTED BY MEANS OF SIMPLIFIED TRIANGLE METHOD FOR SOYBEAN IN PARANÁ STATE – BRAZIL infer the physical limits for solutions of flows of moisture with a minimum of external data. surface energy and the availability of soil Figure 1-. Schematic graph of dispersion of the pixel values of NDVI as a function of surface temperature (Ts) of a satellite image. Source: Adapted from Petropoulos et al (2009), and Ehrlich Lambin (1996), Sandholt et al. (2002) and Nishida et al. (2003). According to Carlson (2013), many agrometeorological model for soybean yield efforts have been invested in complex estimates modify the conventional agro- mathematical methods to obtain surface meteorological model (variable relative parameters such as water content in soil and evapotranspiration - actual evapotranspiration / evapotranspiration; some of these methods potential evapotranspiration, ETr/ETp), by discussed in the literature are unnecessarily spectral variables obtained through the simplified complex. Based on this, Carlson developed a new triangle method to the counties of Campo methodology based on simple and purely Mourao, Luiziana, Roncador, Apucarana, geometric, physical arguments obviating the use Marilândia do Sul, Lidianópolis, Ivaiporã and of complex models. This new methodology is Manoel Ribas, in the years crop. termed herein as "simplified triangle method”. Para Dubreuil et al. (2010), the nature of 2. MATERIALS AND METHODS the spectral behavior of the plants a simple The universe of analysis encompasses relationship between NDVI and the state of Paraná, Campo Mourao, Luiziana, evapotranspiration can be established by means Roncador, Apucarana, Marilândia do Sul, of stomatal resistance to the transfer of water Lidianópolis, Ivaiporã and Manoel Ribas, located vapor. in the southern region of the country between Therefore, it is justified that a good latitudes 22°29'S and 26°43'S and the meridians estimate of crop yield requires modeling the 48°2’W and 54°38'W (Figure 2), where the environmental effects on physiological processes, prevailing climate is temperate mesothermal and determinants of crop yield, a quantity that is not super-humid, Cfa - according to Köeppen climate only an important tool for farmers but also for classification with moderate temperatures, well the food industry and for the planner, leading to distributed rainfall and hot summers. The state is the implementation of appropriate public located in a region of climatic transition from a policies. The aim of this article was to test the subtropical climate with milder north to a 156 FUZZO, D. F. S. SPECTRAL AGROMETEOROLOGICAL MODELING ADAPTED BY MEANS OF SIMPLIFIED TRIANGLE METHOD FOR SOYBEAN IN PARANÁ STATE – BRAZIL condition that approaches the south temperate season of plant growth is better defined. climates, where winters are severe and the Figure 2 - Location of the study area. Source. Silva-Fuzzo, D.F. The counties selected were based on rojection_tool) and were redesigned to WGS-84 data from the SEAB (Department of Agriculture projection and GeoTIFF format. and Supply of State of Paraná). As such, eight In all, 14 NDVI images and 14 Ts images counties that showed high levels of average were composed from 16 and 8 days, respectively. productivity (kg ha-1) in the 10 years analyzed These were therefore selected for the same dates were selected, namely the agricultural years from of satellite for the passage of the two quantities, 2002/2003 to 2011/2012, totaling a period of 10 Ts and NDVI, i.e., the date of the NDVI image is years, with soybean. the same image data for Ts (scenes 257, 273, 289, To estimate relevant parameters using 305, 321, 337 353, 001, 017 033 049 065 081 and the triangle method, images from sensor MODIS, 097, representing respectively the dates, 09/14 products MOD13A2 and MOD11A2, tile h13v11, 09/30 10/16 11/01 11/17 12/03 12/19 01/01 were used. These products are the compositions 01/17 02/02 02/18 03/06 03/22. Were selected of images of 16 days of Vegetation Index NDVI the 14 images of each composition comprising and images of 8 days with surface temperature the agricultural period of soybean growth Ts, respectively, with spatial resolution of one (September to April) totaling 28 images for each kilometer. These images can be obtained from year. Mathematical operations were made by the NASA <https://wist.echo.nasa.gov/api/> site, changing the digital values of the image pixels, being originally in sinusoidal projection and in the MOD11A2 products are for 16-bit images, HDF format (Hierachical Data Format). These and are converted primarily expressed in Kelvin were processed at first by MRTool tool (Modis temperature, and then for values in degrees Reprojection Tool Celsius (°C) through the following equations 1 <https://lpdaac.usgs.gov/lpdaac/tools/modis_rep and 2. 157 FUZZO, D. F. S. SPECTRAL AGROMETEOROLOGICAL MODELING ADAPTED BY MEANS OF SIMPLIFIED TRIANGLE METHOD FOR SOYBEAN IN PARANÁ STATE – BRAZIL ( 1 ) corresponding to the maximum coverage of vegetation (bare soil) (Gillies and Carlson, 1995). (2) According to Carlson, some of these methods discussed in the literature are unnecessary since The MO13A2 products, 16-bit images, the triangle method can be evaluated without were converted to and scale of -1 to 1 by dividing the use of complex models but perhaps with 10,000 by the image, according to equation 3. equal accuracy from simple, purely geometric These procedures were performed using IDL . formulae based on physical arguments. This simplification uses the configuration of the pixel ( 3 ) e n v e l o p e i n T * /Fr space, specifically the boundaries of the envelope, to constrain the With the NDVI values converted from solutions. In so doing, the method’s power is that the digital product and the raw temperatures it does not require any external data for the converted to degrees Celsius, calculations of the atmosphere or the surface. The key parameters important boundary parameters, NDVIs, NDVIo, that define these boundaries are Tsmax, Tsmin, NDVImax, and NDVImin, were performed from NDVIs and NDVIo and the two key features are the NDVI images, where NDVIs and NDVIo are, the warm (dry) and cold (wet) edges, shown in respectively, the highest values represent 100% Figure 1. Thus, the method of the triangle, coverage of vegetation and the lowest values determined from its simple geometric form, represent by bare soil. Similarly, the extraction of yields the availability of surface soil moisture Tmax values are obtained, for the warmest pixels (Mo) and the evapotranspiration (ET) expressed with bare soil or urban area while Tmin as a fraction of the net radiation (Rn), referred to represents the lowest temperatures appropriate as the evapotranspiration fraction (EF), as to areas with well watered dense vegetation as, exemplified in Carlson (2013). Equations 6 and 7 proposed by Carlson, 2007. Subsequent define these terms as follows: scatterplots are shown with axes representing scaled quantities, T* and fractional vegetation (6) cover, Fr, where T* is defined as equation 4 : (4) (7) where, EF is the value appropriate for veg Ts, as defined above, is surface radiant vegetation alone, the potential temperature, Tsmin is the value of Ts for wet evapotranspiration (assumed to be equal to 1.0), bare soil or dense well watered vegetation, and and T* is the scaled surface radiant temperature Tsmax is the corresponding value of Ts for dry, evaluated along the warm edge. Thus. EF Mo soil = bare soil characteristic of an urban center, for and EF = 1 veg example. Fractional vegetation cover is obtained These formula are based on the from NDVI through the equation 5 (Carlson, assumption that the soil surface evaporation and 2007). surface soil moisture availability (Mo) are at maximum (the potential (Mo =1) along the cold (5) edge of the pixel envelope and are zero (Mo = 0) along the warm edge. The transpiration from just the plants is assumed to be always at potential, Here, NDVIo is the NDVI value unless the leaves are wilting. Transpiration and corresponding to bare soil; and NDVIs is the value evaporation are blended together depending on 158 FUZZO, D. F. S. SPECTRAL AGROMETEOROLOGICAL MODELING ADAPTED BY MEANS OF SIMPLIFIED TRIANGLE METHOD FOR SOYBEAN IN PARANÁ STATE – BRAZIL the fractional vegetation cover to yield the total and development, allowing one to monitor the evapotranspiration expressed as EF. water storage of the soil by means of the The original agrometeorological model principle of conservation of mass in a volume of comes from the multiplicative model based on vegetated soil (Pereira et al., 2002). The Doorenbos and Kassam (1979), as proposed by estimated water balance is based on the method Rao et al. (1988). This model (Equation 8) of Thornthwaite and Mather (1955), in decendial considers the product of the ratio, ETr / ETp, scale, which provides estimates of actual reducing production as the water requirement of evapotranspiration (ETR), the water storage in soybean ceases to be met during the the soil (ARM), the water deficit (DEF) and water phenological stages considered, in order that: surplus (EXC). Subsequently, the estimated values of (8) rainfall from the TRMM satellite (Tropical Rainfall Measuring Mission) were also obtained, as well as the average air temperature from the Global Atmospheric Model ECMWF (European Where, Ya is the estimated yield (kg ha- Centre for Medium-Range Weather Forecasts), 1); Yp, the potential yield (kg ha-1); ETr, the actual because this large data set allows for increased evapotranspiration (mm); ETp, the potential coverage of the county, for the years from evapotranspiration (mm); and kyi is the 2002/03 to 2010/11 were obtained for Campo coefficient of penalization productivity due to Mourão and Apucarana city for data comparison. water deficiency for each developmental stage. The TRMM data are available for free online Phenological stage (ky) are, Vegetative access through the NASA website: Development = 0.2, Flowering = 0.4, Grain Filling <http://trmm.gsfc.nasa.gov/data_dir/data.html>. = 0.8, Maturation = 0.2. It can be seen that ETr / The data comes in xls format, geographic ETp, being a relative evapotranspiration, is very coordinates, and temperature data obtained similar to the evapotranspiration fraction EF, from ECMWF air. Images are available free of which is estimated by the triangle method. charge in raster format (tif geo) in the JRC site, Typically, in agrometeorological modeling, the http: //mars.jrc.ec.europa .eu / mars / Aboutus / ETr and ETp values are obtained with the FOODSEC / Data-Distribution. calculation of climatic water balance, with data The obtaining agro-meteorological data obtained from conventional meteorological can be made by networks of meteorological stations. Here, were test the use of EF in place of stations which record atmospheric data. this ratio by replacing the ETr/ETp values in the However, given the size of the country, there is agrometeorological model with EF. not a network of stations with sufficient coverage Thus, the ETr / ETp were replaced by the to meet this need, especially at the local level. value of EF, estimated by the simplified triangle Thus it is necessary to use data estimated by method, consequently the model was modified meteorological satellites such as TRMM for according to Eq.9. Where, EF is fractional precipitation and ECMWF to the air temperature. evapotranspiration. Sowing dates were obtained according to Johann (2012), who estimated date of planting (9) soybeans for the State of Paraná, considering the type of farming, the best conditions of water service in the phenological phases, and showing the regions with higher or lower weather risk for The Water Balance aims to calculate the the development of this vegetation. water storage in the soil taking into account both the type of vegetation and its stage of growth 159 FUZZO, D. F. S. SPECTRAL AGROMETEOROLOGICAL MODELING ADAPTED BY MEANS OF SIMPLIFIED TRIANGLE METHOD FOR SOYBEAN IN PARANÁ STATE – BRAZIL To evaluate the reliability of the data To investigate the method, the values estimated by agrometeorological model, first the estimated by the modified agrometeorological multitemporal compositions were made in RGB model were compared using data from fractional images of NDVI, to highlight only a summer crop evapotranspiration (EF), with the conventional and extract the pixels of an urban area. Thus, the model using the values of relative period with the highest vigor were placed on the evapotranspiration (the radio of actual R channel, then the images with less vigor in G evapotranspiration to potential and B channels. This procedure was evapotranspiration (ETr/ETp) obtained by the operationalized by 4.5 ENVI (The Environment for climatological water balance of Thornthwaite and Visualizing Images), and the composition of these Mather (1955) for estimated data TRMM and scenes were the image 001 (refers to January, ECMWF. In order to the satellite image data be 1st), image 285 (refers to October, 16th), image compatible with the crop measurements, satellite 305 (refers to November, 1st), the year 2010/11 estimates of EF for individual pixels were were used. averaged over each county. And with the aim to separate the pixels Statistical analyzes were performed to for summer crops from other crops, a procedure verify the correctness and accuracy of the data, was developed in IDL (Interactive Data Language) using a simple linear regression model and its the extraction system data / pixel of RGB coefficient of determination (R2), which shows composite image in grayscale, i.e. through that the ratio or percentage of variance in one system were chosen the gray values between 0 variable that can be explained variance from the and 255, in which for the R channel were other. It is noteworthy that the accuracy refers to considered the largest gray level values and G the degree of conformity of an estimated and B channel values lower levels of gray. Thus, quantity to the true (measured directly in the the pixels are extracted (removed) at lower gray field) value. The precision is the degree of values, creating the mask for the summer crop. variation in the results of a measurement and is These were taken as the basis for comparison of based on the standard deviation of a number of LANDSAT 5/ TM images mosaics as ground repetitive values generated from the same reference. analysis. The MAE (mean absolute error) may be In the making of masks, were separated most appropriate for checking the correctness or areas with summer crops (soybeans and corn) accuracy of estimated scalar data (e) in relation from the other cover crops. Among the various to the (measured) data (o) (Eq.10). The RMSE simulations of cuts of gray levels, for channels of (root square error) is used to estimate the quality RGB color composition, it was found that the one of the classifier, as in Eq. 11. with the best masking properties was obtained with R-180 and GB -110 for crop years, i.e., were (10) selected pixels with gray levels > 180 in R and gray levels < 110 in GB channels . So, did not include urban areas on the agro-meteorological model estimated harvest. Later, the average was (11) calculated from the pixels obtained by vegetation mask for each of the 8 counties, (pixels estimated To verify the final quality of the by agrometeorological model with productivity estimator model, as described by Willmott et al. values). This was necessary because the (1985) propose an adaptation called index productivity values observed in the field, Willmott modified, expressed in Eq.12, given by: corresponds a total value for the entire county area. 160 FUZZO, D. F. S. SPECTRAL AGROMETEOROLOGICAL MODELING ADAPTED BY MEANS OF SIMPLIFIED TRIANGLE METHOD FOR SOYBEAN IN PARANÁ STATE – BRAZIL (12) typical triangular configuration the dashed red line indicating the warm (dry) edge of the triangle, and the dotted line in blue the cold (wet) where d1 is the concordance index and the edge. Hot and cold ends, respectively, subscripts are defined as in Eq. 10. correspond to the driest and wettest pixels for a The confidence index "c" indicates the given value of Fr (Jiang et al, 2001; Petropoulos et model performance according to Sentelhas and al, 2009; Garcia et al, 2014). Camargo (1997) through the index of precision These scatterplots can be interpreted in and accuracy, expressed by the Eq. 13. the context of Eqs 6 and 7 as follows: Each pixel within these triangles represents a different value C = R² *d ( 1 3 ) of surface soil water content (Mo) and evapotranspiration fraction (EF). The former This criterion was also applied to the varies linearly (along any straight line at constant data set, the Mann-Whitney (1947). A criterion of Fr) from zero (dry) at the warm edge (the sloping 5% or less is used, in order to ascertain whether red line) to 1.0 (wet) at the cold edge (the vertical there was any significant difference between the blue line). EF varies linearly from zero at the mean and the distribution of the measured crop lower right hand vertex to 1.0 along the cold productivity data and those derived from the edge. Fractional vegetation cover varies linearly simplified triangle method. In addition, the from zero at the base of the triangle to 1.0 at the average systematic error (Es) and the (non- upper vertex. systematic) random mean error (Ea) were In Figure 3, the pixels tend to form a calculated. more dense grouping on top of the scatter plot in the months of higher vegetative peak of 3. RESULTS AND DISCUSSION soybeans, whereas in the months of vegetative Estimates of EF and Mo, using the growth more pixels are dispersed within the simplified triangle method, were obtained from triangular shape, as would be expected according the images T * and Fr, in the form of the to the physiological cycle for soybeans. It is also approximately triangular-shaped scatterplots for very important to analyze the upper vertex of the the eight counties of the state of Paraná, for 10 triangle, because this is relevant to the years of crop data (2002/03 to 2011/12). As an distribution of the pixel envelope, the greater the example, the counties of Campo Mourão and the number or pixels accumulated at the top of the agricultural year 2011 -12, show a scattering of triangle, the higher the concentration of pixels for each satellite image, representing the vegetation for that image, ie, during the time entire phenological cycle of soybean. In Figure 3 representing the greatest crop development. were observe the dispersion of pixels in their 161 FUZZO, D. F. S. SPECTRAL AGROMETEOROLOGICAL MODELING ADAPTED BY MEANS OF SIMPLIFIED TRIANGLE METHOD FOR SOYBEAN IN PARANÁ STATE – BRAZIL Figure 3 - Schematic representation of triangle method and phenological cycle of the soybean crop. Source. Silva-Fuzzo, D.F. According to Carlson (2013), the sloping Then were tested the agrometeorological warm side of the triangle (the warm edge) model of Doorenbos and Kassam (1979), which indicates higher bare soil temperature than in includes EF values obtained from the simplified areas of higher vegetation, because the soil is triangle method. Agricultural productivity of the more shielded from the sun by the vegetation. soybean crop was evaluated by considering the Were note with triangular graphs that the warm estimated values of EF using the simplified edge has many more pixels near the lower part of triangle method, and substituting these for the the warm edge (signifying more dry, canopy values of ETr/ETp (ratio between actual visible to the radiometer), for the months with evapotranspiration/potential evapotranspiration) the greatest amount of exposed soil, i.e. the previously used in the original model, whose months of sowing (Sept / Oct) and months input was based on surface, rainfall and climate regarding the harvesting season (Mar / Apr). data as described earlier in this paper). The It is worth pointing out that some model performance was evaluated by means of sensitivity to T* does exist near the upper vertex statistical analyzes presented below, the index " in some of the triangles, indicating the soil is not d " and " d " Willmott, coefficient of 1 2 always invisible to the satellite sensor under the determination Rsquared, index "c", and MAE, dense vegetation. When there is not a group of RMSE, Ea and Es errors, and the Mann-Whitiney pixels at the top of the triangular space, this may (1947) test (Table 1). indicate that there are few patches of dense Table 1 express the differences between vegetation within the triangle space, (Fr) less measured and estimated yield is expressed as a than 1.0. Although the temperature of the leaves percentage 100 (b-a)/a, d are the concordance 1, themselves is assumed not to vary significantly, indices, which indicates the model performance, the temperature of the vegetation canopy (soil R2 is the coefficient of determination, (the plus crop) varies with soil moisture, so that correlation coefficient). The parameter “c” is a months with less vegetation have a greater confidence index; it is a measure of the model exposure to bare soil. Recall, that the basic performance. MAE is the mean absolute error, assumption in the triangle method is that the RMSE is the root squared error, Ea is the temperature of the leaves does not vary in space systematic error, Es is the non-systematic error, unless the crop is experiencing significant water and the p-value is an assessment of the stress. significance significant difference between the 162 FUZZO, D. F. S. SPECTRAL AGROMETEOROLOGICAL MODELING ADAPTED BY MEANS OF SIMPLIFIED TRIANGLE METHOD FOR SOYBEAN IN PARANÁ STATE – BRAZIL mean values of the estimated data and the mean defined as the difference between estimated and values of the measured data. Error here is measured crop productivity. Table 1 - Statistical Analysis performance of the agro-meteorological model (Eq. 9) modified for simplified triangle method, from 2002/03 to 2011/12 in kg ha-1, and data measured by SEAB. Note which “Obs” means observed productivity and “Est” estimated productivity. *Significant at 5% for the p-values. By analyzing the coefficients of Apucarana and Ivaiporã showed low values of confidence "c", the values ranged from 0.9 this index 0.53 and 0.55, respectively, indicating (c>0.85) for some counties as Campo Mourão. the poor performance of these models, for these 163