Agricultural drought in Chile From the assessment toward prediction using satellite data.
Loading...
Date
2017
Journal Title
Journal ISSN
Volume Title
Publisher
Universidad de Concepción
Abstract
Climate change is occurring and there is a scientific consensus that human being is playing a key role by pouring greenhouses gases to the atmosphere. Temperature has been increasing globally and the precipitation patterns are changing. Regionally, since the year 2010 Chile has been experiencing which has been called a mega drought, however, it has been seen mostly in meteorological terms by analyzing precipitation deficits. Further, the future projection for Chile indicates that the precipitation will decrease in Central South Chile, this addded to the increase on temperature likely could increase drought frequency and intensity. Also, in this regard crop yield of corn and wheat decreases are forecasted by 2050 for Chile.
The study on how climate variability and human activity impact agriculture has been known as agricultural drought. One of the main factors that trigger this drought conditions is precipitation deficit, thus is crucial to understand how this depletion relates to agriculture development. Although, since 2010 Chile has been facing water shortage mostly as results of the analysis of annual precipitation, but still there is a lack of knowledge about how this mega drought is affecting agriculture over Chile. Moreover, during the growing season 2007-2008 a large part of the country experienced decreases in crop yield for which these areas were declared under drought emergency by the government. However, by analyzing the total amount of annual precipitation these years are not seen as relevant drought years. This happens in part because for vegetation is more important the timing of the rainfall deficit rather than the cumulative over a year. Thus, the study and understanding of agricultural drought and methods that could help to anticipate it are challenging.
The study of agricultural drought at regional and global scale brings the problem of having enough data that allow to analyze it spatially and temporally. Nonetheless, since the 70’s the use of remote sensing data obtained from satellite to monitor the environment at global and regional scale has been highly improved, and nowadays are a key data source to support climatic and environmental studies. In that regard, there is an important amount of satellite-derived data publicly available. One of this dataset that provided useful data for the monitoring of vegetation is provided for the National Aeronautics and Space Administration (NASA) and its sensor the Moderate-Resolution Imaging Spectroradiometer (MODIS) which is coupled to the TERRA and AQUA satellites. Further, multiple microwave and infrared satellites have allowed the development of precipitation estimates products at different temporal and spatial resolutions Between them, highlight the Tropical Rainfall Measuring Mission (TRMM) Multi-satellite Precipitation Analysis (TMPA) having data since 1998, and also has been derived long-term precipitation products such as the Precipitation Estimation from Remotely Sensed Information using Artificial Neural Networks - Climate Data Record (PERSIANNCDR) with data since 1983, and the Climate Hazards Group InfraRed Precipitation with Station data version 2 (CHIRPS v2) providing estimates since 1981. These vegetation and precipitation satellite products are valuable data sources for agricultural drought studies, allowing to evaluate the interaction vegetationprecipitation at regional and global scale.
Accordingly, in this thesis was studied the usefulness of satellite data for the assessment and prediction of agricultural drought over Chile. The main research question is: How well the satellite data of vegetation and precipitation together with climatic oscillation indices can be used to predict agricultural drought before the end of the growing season? To achieve this, the work was developed in three stages: 1) assessment of vegetation response to water shortage in the BioBío Region of Chile, 2) the evaluation of long-term satellite precipitation data over Chile for use in drought studies, and 3) prediction of agricultural drought in Chile from one to four month before the end of the growing season for 2000-2016. For the first stage, was used the Vegetation Condition Index (VCI) which was derived from the Normalized Difference Vegetation Index (NDVI) provided by the MODIS Vegetation Indices (VI) product MCD12Q1.005. The land cover product MCD12Q2.051 from MODIS was used to derive a cropland mask for the BioBio region. Besides, twenty six weather stations were used to derive the Standardized Precipitation Index (SPI) at time-scales from one to six months. To understand how the vegetation response to water shortage, the VCI over the growing season was correlated against the SPIs. For this the data was aggregated for the whole region and per administrative unit, and also was analyzed the point-to-point correlation. For the second stage, were used 758 weather station over Chile having monthly precipitation data since 1981 to 2015. Then, were selected the long-term satellite data products, CHIRPS v2 for 1981-2015, and the PERSIANN-CDR for 1983-2015. The accuracy when measuring monthly precipitation was evaluated for these two long-term precipitation products together with the TMPA product. These three products were compared with the point precipitation data obtained at weather stations. Also, the long-term products were evaluated for drought monitoring using the SPI for one, three, and six months. Finally, for the third stage, the study area was defined for 29-41°S through Chile. Over this region, the product MCD12Q2.051 from MODIS was used to derive a single cropland mask over Chile.
From the 2221 census unit used for the Ministry of Agriculture of Chile for the 10-year agricultural census over the study area, 758 were selected by filtering using the cropland mask. Per census unit was extracted the growing season start (SOS) and end (EOS) from the product MCD12Q2.005 from MODIS. Over each census unit was calculated a proxy of biomass production, the anomaly of cumulative NDVI (zcNDVI) for 2000-2016. As predictors per census unit were calculated from one to four months before EOS the zcNDVI and SPI at one, three, six, twelve, and twenty-four months. Also, were used as a predictor for one to four months before EOS the three-month average Pacifical Decadal Oscillation (PDO) and Multivariate ENSO index (MEI), which were also lagged at 0, 3, and 6 months; from the corresponding prediction timing months before EOS. Two methods were assessed for the prediction, an Optimal Linear Regression (OLR) per census unit which selects the predictor that produce less error with a cross-validated linear regression. The second method corresponds to a Multi Layer Feed-Forward Neural Network also called Deep Learning (DL), for which were created cross-validated models considering all units and using as additionals predictor latitude and longitude.
Results from the first stage showed that the 3-month SPI (SPI-3), calculated for the modified growing season (Nov-Apr) instead of the regular growing season (Sept-Apr), has the best Pearson correlation with VCI values with an overall correlation of 0.63 and between 0.40 and 0.78 for the administrative units. These results show a very short-term vegetation response to rainfall deficit in September, which is reflected in the vegetation in November, and also explains to a large degree the variation in vegetation stress. It is shown that for the last 16 years in the BioBío Region we could identify the 2007 2008, 2008-2009, and 2014-2015 seasons as the three most important drought events; this is reflected in both the overall regional and administrative unit analyses. These results concur with drought emergencies declared by the regional government. Next, from the second stage, results showed that the monthly analysis for all satellite products highly overestimated rainfall in the arid North zone. However, there were no major differences between all three products from North to South-Central zones. Further, in the South zone, PERSIANN-CDR shows the lowest fit with high underestimation, while CHIRPS 2.0 and TMPA 3B43 v7 had better agreement with in-situ measurements.
The accuracy of satellite products were highly dependent on the amount of monthly rainfall with the best results found during winter seasons and in zones (Central to South) with higher amounts of precipitation. PERSIANN-CDR and CHIRPS 2.0 were used to derive SPI at time-scales of 1, 3 and 6 months, both satellite products presented similar results when were compared in-situ against satellite SPIs. Because of its higher spatial resolution that allows better characterizing of spatial variation in precipitation pattern, the CHIRPS 2.0 was used to mapping the SPI-3 over Chile. Finally, from the third stage, results from the two prediction models evaluated (OLR and DL) showed similar and good prediction accuracy, with mean R2 cv values for OLR of 0.94, 0.79, 0.63 and 0.51, and for DL of 0.93, 0.79, 0.63 and 0.51, for one, two, three and four months before EOS respectively. Also, was discussed potential model improvements and how the method could contribute to an early warning system for agricultural drought in Chile.
Description
Tesis presentada para optar el grado de Doctor en Ingeniería Agrícola con mención en Recursos Hídricos en la Agricultura
Keywords
Sequía - Chile, Cambios de la Temperatura Global, Calentamiento Global, Agricultura - Efecto del Calentamiento Global