El objetivo del trabajo fue determinar y comparar la capacidad de distintos índices (climáticos y espectrales) de estimar los rendimientos de maíz y soja a escala departamental en la provincia de Entre Ríos.
Comparación de índices climáticos y espectrales en la estimación de rendimiento de maíz y soja a nivel departamental en Entre Ríos
Tesis presentada para optar al título de Magíster de la Universidad de Buenos Aires en Meteorología Agrícola
- Maestrando: Ing. Agr. Nicolás Vaiman
- Director de Tesis: M. Sc. Roberto De Ruyver
- Co-Director: Dr. Martín Durante
- Buenos Aires, 19 de junio de 2018
Las estimaciones de rendimientos agrícolas son de gran importancia debido a que la producción de alimentos juega un papel fundamental en la seguridad alimentaria y en la economía de los países. En el pasado, las estimaciones de rendimiento se debían realizar a campo, con operaciones trabajosas y lentas, y se obtenían resultados poco precisos. Por este motivo, cada vez más trabajos apuntan a estimar el rendimiento mediante técnicas geo-informáticas. El objetivo del trabajo fue determinar y comparar la capacidad de distintos índices (climáticos y espectrales) de estimar los rendimientos de maíz y soja a escala departamental en la provincia de Entre Ríos.
Por otro lado, se evaluó la capacidad de estimar precipitaciones con datos de TRMM (Tropical Rainfall Measurement Mission), es decir, precipitaciones derivadas de satélites. En este sentido se determinó que, a nivel mensual, estos datos son más precisos para estimar la precipitación real si se los compara con los métodos clásicos de interpolación. Por otro lado, se determinó que los datos de TRMM acumulados cada 15 días son adecuados para diversos usos agronómicos.
Para estimar rendimientos, tanto en maíz como en soja, las mejores estimaciones de rendimiento se lograron dos meses previos a la cosecha con modelos derivados de índices espectrales. El mejor modelo presentó un error de 363 kg.ha-1 en maíz y de 132 kg.ha-1 en soja. Los modelos lineales obtenidos en este trabajo aportarían: 1) objetividad en las estimaciones, 2) adecuada anticipación en las estimaciones y 3) datos a nivel departamental.
Climatic and spectral indices comparison in the assessment of maize and soybean yields at departmental scale in Entre Ríos
The estimates of agricultural yields are of great importance because food production plays a fundamental role in food security and the economy of the countries. In the past, yield estimates had to be made in the field, with laborious and slow operations, and imprecise results were obtained. For this reason, more and more studies aim to estimate yields through geo-informatics. Considering the availability of free access data there are two possibilities to estimate the yields: on the one hand, precipitation data, and, on the other hand, data from remote sensing can be used. The objective of this work was to determine and compare the capacity of different indices (climatic and spectral ones) to estimate agricultural yields at departmental scale in the province of Entre Ríos. In the first instance, the capacity of the precipitations and the SPI (Standardized Precipitation Index) of 3 months were determined, both in different periods of accumulation in the warm semester (October to March). The models that best explained the maize yields were those that included December information (both with precipitation and with SPI), with a lower prediction error and a higher coefficient of determination. In contrast, for soybeans the best models were those that were based on precipitation and SPI prior to the critical period. Then, in the second instance, the capacity of TRMM (Tropical Rainfall Measurement Mission) to estimate precipitation in Entre Ríos was evaluated. In a first monthly analysis, the estimate derived from TRMM and two interpolation methods were compared to nine conventional meteorological stations (CME) data (five inside and four outside the province). In a second analysis for different accumulation periods less than a month, three CME within the province were used. On a monthly basis, the ability to estimate precipitation was better with TRMM than with either of the two interpolation methods. The estimation of monthly precipitation with the equation with all grouped data of TRMM was not different from the particular estimation of each CME. That allows us to use the equation in a generalized manner for the
whole province. On a scale smaller to the month, it was observed that the adjustment increases with the accumulated days. The 15-day accumulation period was established as the minimum that does not affect the accuracy with respect to longer periods. Therefore, the TRMM data accumulated every 16 days (frequency of MODIS products) were used in conjunction with other spectral indices in the following instance. In this third instance, the capacity of a set of spectral indices to estimate yields was determined. For this, reflectance values of different MOD09A1 product bands and the surface temperature of the MOD11A2 product were used, both belonging to MODIS. The resulting indices (basic and scaled every 8 days, and combined every 16 days), were averaged over 8 time periods in different phenological stages of the crop cycle. The most accurate models to estimate soybean and maize yields were those whose indices had been scaled. In maize, the best indices were obtained when considering only the critical period. In contrast, for soybeans, the best models included the critical period of the crop and the full cycle. In general, the best results were obtained when using the NDWI and NDDI (derived from the NDWI). Ultimately, climatic and remote sensing indices were compared in the yield estimation. In maize, the best estimate with climatic indices was obtained with the SPI of January and February in Federal, with an RMSE of 820 kg.ha-1 (relative error of 22%). On the other hand, the best estimate with spectral indices was obtained with the Sc_NDWI_7 in the critical period in Villaguay, with an RMSE of 363 kg.ha-1 (relative error of 8%). In soybean, the best estimate with climatic indices was obtained with the SPI from January to March in the Federación, with an RMSE of 245 kg.ha-1 (relative error of 8%). In contrast, the best estimate with spectral indices was obtained with the Sc_NDWI_7 in the period between January 17 and February 18 in Nogoyá, with an RMSE of 132 kg.ha-1 (relative error of 5%). The results of this work indicate that the estimated precipitations with TRMM could be used for agronomic uses. In addition, good estimates of departmental yields were obtained through models from simple linear regressions. In both maize and soybeans, the best yield estimates were achieved two months before harvest. The linear models obtained in this work could be expanded with more data in future years or could be operationally implemented in its original form. These models would provide: 1) objectivity in the estimates, 2) adequate anticipation in the estimates and 3) data at the departmental level.
Keywords: yield estimate, soybean, maize, precipitations, remote sensing