The VegDRI is a hybrid drought index that incorporates multiple sources of input data informative of drought stress. The VegDRI methodology includes many steps shown in Figure 1 and our publications can be consulted for further technical details.
Steps A, B, and C form the model and data development phase. Steps D, E, and F comprise the weekly operational map production phase. Step A shows many input satellite, climate, and biophysical data layers. Step B outlines the extraction and assembly of a large historical database of all input data designed for VegDRI model training. In Step C, the training data are input into regression tree analysis to derive multiple models through time during overlapping 8-week periods. Each model can then be used to estimate drought conditions for a certain time of year based on the condition of the dependent variable, the PDSI. In Step D, all input geospatial data layers are assembled. In Step E, the operational VegDRI system software is implemented where the VegDRI model is applied to the geospatial input data. And finally, in Step F, the VegDRI map is output in various image and graphic formats and distributed through project websites and map viewers.
Step A
Step A shows many input satellite, climate, and biophysical data layers. Satellite data types include MODIS Percent of Average Seasonal Greenness (PASG) and the Start of Season Anomaly (SOSA). Climate variables include the Palmer Drought Severity Index (PDSI) and the Standardized Precipitation Index (SPI). Five biophysical data layers provide the environmental and geographic setting for the climate and satellite vegetation conditions. They include land use/land cover type, irrigated agricultural land, soil available water capacity, elevation, and ecological region.
The PASG is an indicator of accumulated greenness (or NDVI) during the growing season compared to the historical average for the same point in time. The amount is expressed as a percent of the historical average. A PASG value of 100% means that the current seasonal greenness is equal to the average historical greenness and this is indicative of normal or average vegetation conditions. PASG values less than 100% indicate below average greenness (poorer than normal vegetation conditions) that may be linked to some form of stress (for example, drought, flooding, fire, hail damage, or pest infestation). PASG values greater than 100% indicate higher than average greenness, which reflect above normal vegetation conditions. PASG values are not calculated for a location until the start of season has occurred.
The SOSA is a measure of the temporal difference in the start of the growing season (SOS) for a given year compared to the historical average SOS for a location. SOS can be defined as the time when vegetation initiates growth (that is, photosynthetic activity) after winter as observed from satellite observations. A negative SOSA indicates that the SOS for a specific year is earlier than the average date and a positive SOSA appears when green-up occurs later than normal. The SOSA is used to express the changes in vegetation phenology that may occur from year-to-year.
Two climate variables in VegDRI indicate dry conditions in a meteorological sense, the PDSI and the SPI. The SPI was designed to quantify precipitation differences over multiple time intervals (for example, 1- to 12-month periods) based on fitting a long-term precipitation record at a given location over a specified interval to a probability distribution, which is then transformed into a gamma distribution such that the mean SPI value for that location and time period is zero. SPI values are positive if the precipitation over a specific time period is higher than the historical average precipitation over that same period and negative if precipitation is less than the historical mean.
The PDSI is a prominent drought index that has been widely used to assess agricultural drought. The PDSI is calculated using a simple supply-and-demand water balance model that integrates precipitation and temperature data, as well as the available water holding capacity of the soil. A self-calibrated PDSI calculation is used in VegDRI. This version of PDSI utilizes a mathematical calibration of model coefficients tuned to local climate and soil characteristics. This essentially improves the spatial comparability of PDSI values and calibrates the index so that extreme dry and wet events have a comparable rate of occurrence across the nation. The PDSI is used in VegDRI models as the dependent variable.
The biophysical data layers help to differentiate drought stress by giving information on environmental and geographic context. The land use/land cover (LULC) type represents broad categories of vegetation with various seasonal cycles and climate-vegetation responses. An irrigated agriculture variable differentiates irrigated lands, which are less susceptible to drought stress because of targeted water applications, from rainfed agriculture. The available water capacity (AWC) variable represents the potential of the soil to hold moisture that is available to plants, which in turn influences the sensitivity of vegetation to drought stress. Elevation data provides different gradients that change the typical seasonality of vegetation. Ecological regions (that is, ecoregions) give a wall-to-wall geographic framework that accounts for the wide variability in environmental conditions encountered across the country that can influence sensitivity and responsiveness of vegetation to drought stress. Ecoregions typically have similar ecosystems and environmental resources defined using both abiotic (for example, physiography) and biotic (for example, plant species) criteria.
Step B
n Step B a training database of all climate, satellite, and biophysical data was extracted and assembled in order to train VegDRI models. Historical, point-based PDSI and SPI data records at over 2,000 weather stations first populated the VegDRI training database. Training data were also extracted from the satellite and biophysical data (that are in the form of image grids) at the location of each weather station. A 3-by-3 pixel window centered on each station location was used to calculate the average value across all pixels in the window for continuous variables (for example, PASG) and the majority value for categorical variables (for example, LULC). Data values associated with both urban and water locations were excluded. The biophysical variable values calculated for each station were held static across the historical period. Historical records of all stations in the database were then temporally subset into 26 bi-weekly periods (e.g., bi-weekly period 1: January 1 to 14) across the calendar year to train a series of separate, bi-weekly VegDRI models.
Step C
Step C encompasses VegDRI model development. For each period, a commercial classification and regression tree (CART) algorithm called Cubist was used to analyze the historical data in the training database and generate a rule-based, piecewise linear regression VegDRI model. Each model incorporates historical data for the ‘dynamic’ climate and satellite-based variables while holding the biophysical variables constant over an 8-week window that includes the current period and extends to the 7 prior weeks. As discussed earlier, the self-calibrated PDSI serves as the dependent variable in these empirical-based models, providing a well-established classification system for VegDRI to categorize varying levels of drought severity on vegetation based on the analysis of the other biophysical, climate, and satellite variables. Twenty-six period-specific VegDRI models were developed through the year spaced at every two weeks.
Each VegDRI regression tree model consists of an unordered set of rules with each rule having the syntax ‘if x conditions are met then use the associated linear regression model’ to calculate a VegDRI value. The following is an example of one of many rules in a single model:
Rule1:
if: LandCover in {Grassland, Pasture/Hay, Row Crops}
Ecoregion in {Western High Plains, Central Great Plains}
36-week SPI ≤ -1.4
AWC ≤ 4.5
PASG ≤ 50
then: VegDRI = -3.5 + 0.6 PASG + 1.48 SPI – 0.14 AWC + 0.25 Percent Irrigated.
Step D
In Step D, the gridded input data layers listed in Step A are assembled together for operational map production. Each data layer has a common spatial domain and grid cell size. Step D also consists of near real time processing of the dynamic satellite and climate data input grids. The satellite source (MODIS) data are processed through a system called eMODIS (expedited MODIS) that provides imagery less than 24 hours later than the last observation in a weekly composite. Additional processing steps such as temporal smoothing and mathematical calculations are necessary to create the PASG and SOSA variables for input into the step E. The gridded climate data (SPI) are calculated from daily precipitation observations acquired from the Applied Climate Information System (ACIS; https://www.rcc-acis.org) and then an interpolated continuous grid is estimated using a method called inverse weighted difference.
Step E
In Step E, software code called MapCubist is used to apply the model rules for each specific period to data in a geospatial format. This is the main step for creating VegDRI map output. In this step, model rules that were developed in Step C are applied to each grid cell in the VegDRI spatial domain. In other words, if the data associated with a grid cell meet the threshold criteria for the three continuous variables and are represented by one of the three land cover types and in either ecoregion, then the lhttps://vegdri.unl.edu/inear regression equation is used to calculate a VegDRI value. Each independent climate, satellite, and biophysical variable is incorporated into a subset of the multiple rules and regression equations that are collectively utilized to calculate the final VegDRI value for each grid cell.
Step F
Finally, Step F consists of distributing data to collaborators and other users, generating a period-specific map pdf, and loading the data into a database for viewing in project map viewers. The USGS viewer (https://vegdri.cr.usgs.gov/viewer/) offers interactive viewing and web services. Pregenerated national, regional, and state maps are also offered at the National Drought Mitigation Center VegDRI web site (https://vegdri.unl.edu/).
The VegDRI is a hybrid drought index that incorporates multiple sources of input data informative of drought stress. The VegDRI methodology includes many steps shown in Figure 1 and our publications can be consulted for further technical details.
Steps A, B, and C form the model and data development phase. Steps D, E, and F comprise the weekly operational map production phase. Step A shows many input satellite, climate, and biophysical data layers. Step B outlines the extraction and assembly of a large historical database of all input data designed for VegDRI model training. In Step C, the training data are input into regression tree analysis to derive multiple models through time during overlapping 8-week periods. Each model can then be used to estimate drought conditions for a certain time of year based on the condition of the dependent variable, the PDSI. In Step D, all input geospatial data layers are assembled. In Step E, the operational VegDRI system software is implemented where the VegDRI model is applied to the geospatial input data. And finally, in Step F, the VegDRI map is output in various image and graphic formats and distributed through project websites and map viewers.
Step A
Step A shows many input satellite, climate, and biophysical data layers. Satellite data types include MODIS Percent of Average Seasonal Greenness (PASG) and the Start of Season Anomaly (SOSA). Climate variables include the Palmer Drought Severity Index (PDSI) and the Standardized Precipitation Index (SPI). Five biophysical data layers provide the environmental and geographic setting for the climate and satellite vegetation conditions. They include land use/land cover type, irrigated agricultural land, soil available water capacity, elevation, and ecological region.
The PASG is an indicator of accumulated greenness (or NDVI) during the growing season compared to the historical average for the same point in time. The amount is expressed as a percent of the historical average. A PASG value of 100% means that the current seasonal greenness is equal to the average historical greenness and this is indicative of normal or average vegetation conditions. PASG values less than 100% indicate below average greenness (poorer than normal vegetation conditions) that may be linked to some form of stress (for example, drought, flooding, fire, hail damage, or pest infestation). PASG values greater than 100% indicate higher than average greenness, which reflect above normal vegetation conditions. PASG values are not calculated for a location until the start of season has occurred.
The SOSA is a measure of the temporal difference in the start of the growing season (SOS) for a given year compared to the historical average SOS for a location. SOS can be defined as the time when vegetation initiates growth (that is, photosynthetic activity) after winter as observed from satellite observations. A negative SOSA indicates that the SOS for a specific year is earlier than the average date and a positive SOSA appears when green-up occurs later than normal. The SOSA is used to express the changes in vegetation phenology that may occur from year-to-year.
Two climate variables in VegDRI indicate dry conditions in a meteorological sense, the PDSI and the SPI. The SPI was designed to quantify precipitation differences over multiple time intervals (for example, 1- to 12-month periods) based on fitting a long-term precipitation record at a given location over a specified interval to a probability distribution, which is then transformed into a gamma distribution such that the mean SPI value for that location and time period is zero. SPI values are positive if the precipitation over a specific time period is higher than the historical average precipitation over that same period and negative if precipitation is less than the historical mean.
The PDSI is a prominent drought index that has been widely used to assess agricultural drought. The PDSI is calculated using a simple supply-and-demand water balance model that integrates precipitation and temperature data, as well as the available water holding capacity of the soil. A self-calibrated PDSI calculation is used in VegDRI. This version of PDSI utilizes a mathematical calibration of model coefficients tuned to local climate and soil characteristics. This essentially improves the spatial comparability of PDSI values and calibrates the index so that extreme dry and wet events have a comparable rate of occurrence across the nation. The PDSI is used in VegDRI models as the dependent variable.
The biophysical data layers help to differentiate drought stress by giving information on environmental and geographic context. The land use/land cover (LULC) type represents broad categories of vegetation with various seasonal cycles and climate-vegetation responses. An irrigated agriculture variable differentiates irrigated lands, which are less susceptible to drought stress because of targeted water applications, from rainfed agriculture. The available water capacity (AWC) variable represents the potential of the soil to hold moisture that is available to plants, which in turn influences the sensitivity of vegetation to drought stress. Elevation data provides different gradients that change the typical seasonality of vegetation. Ecological regions (that is, ecoregions) give a wall-to-wall geographic framework that accounts for the wide variability in environmental conditions encountered across the country that can influence sensitivity and responsiveness of vegetation to drought stress. Ecoregions typically have similar ecosystems and environmental resources defined using both abiotic (for example, physiography) and biotic (for example, plant species) criteria.
Step B
n Step B a training database of all climate, satellite, and biophysical data was extracted and assembled in order to train VegDRI models. Historical, point-based PDSI and SPI data records at over 2,000 weather stations first populated the VegDRI training database. Training data were also extracted from the satellite and biophysical data (that are in the form of image grids) at the location of each weather station. A 3-by-3 pixel window centered on each station location was used to calculate the average value across all pixels in the window for continuous variables (for example, PASG) and the majority value for categorical variables (for example, LULC). Data values associated with both urban and water locations were excluded. The biophysical variable values calculated for each station were held static across the historical period. Historical records of all stations in the database were then temporally subset into 26 bi-weekly periods (e.g., bi-weekly period 1: January 1 to 14) across the calendar year to train a series of separate, bi-weekly VegDRI models.
Step C
Step C encompasses VegDRI model development. For each period, a commercial classification and regression tree (CART) algorithm called Cubist was used to analyze the historical data in the training database and generate a rule-based, piecewise linear regression VegDRI model. Each model incorporates historical data for the ‘dynamic’ climate and satellite-based variables while holding the biophysical variables constant over an 8-week window that includes the current period and extends to the 7 prior weeks. As discussed earlier, the self-calibrated PDSI serves as the dependent variable in these empirical-based models, providing a well-established classification system for VegDRI to categorize varying levels of drought severity on vegetation based on the analysis of the other biophysical, climate, and satellite variables. Twenty-six period-specific VegDRI models were developed through the year spaced at every two weeks.
Each VegDRI regression tree model consists of an unordered set of rules with each rule having the syntax ‘if x conditions are met then use the associated linear regression model’ to calculate a VegDRI value. The following is an example of one of many rules in a single model:
Rule1:
if: LandCover in {Grassland, Pasture/Hay, Row Crops}
Ecoregion in {Western High Plains, Central Great Plains}
36-week SPI ≤ -1.4
AWC ≤ 4.5
PASG ≤ 50
then: VegDRI = -3.5 + 0.6 PASG + 1.48 SPI – 0.14 AWC + 0.25 Percent Irrigated.
Step D
In Step D, the gridded input data layers listed in Step A are assembled together for operational map production. Each data layer has a common spatial domain and grid cell size. Step D also consists of near real time processing of the dynamic satellite and climate data input grids. The satellite source (MODIS) data are processed through a system called eMODIS (expedited MODIS) that provides imagery less than 24 hours later than the last observation in a weekly composite. Additional processing steps such as temporal smoothing and mathematical calculations are necessary to create the PASG and SOSA variables for input into the step E. The gridded climate data (SPI) are calculated from daily precipitation observations acquired from the Applied Climate Information System (ACIS; https://www.rcc-acis.org) and then an interpolated continuous grid is estimated using a method called inverse weighted difference.
Step E
In Step E, software code called MapCubist is used to apply the model rules for each specific period to data in a geospatial format. This is the main step for creating VegDRI map output. In this step, model rules that were developed in Step C are applied to each grid cell in the VegDRI spatial domain. In other words, if the data associated with a grid cell meet the threshold criteria for the three continuous variables and are represented by one of the three land cover types and in either ecoregion, then the lhttps://vegdri.unl.edu/inear regression equation is used to calculate a VegDRI value. Each independent climate, satellite, and biophysical variable is incorporated into a subset of the multiple rules and regression equations that are collectively utilized to calculate the final VegDRI value for each grid cell.
Step F
Finally, Step F consists of distributing data to collaborators and other users, generating a period-specific map pdf, and loading the data into a database for viewing in project map viewers. The USGS viewer (https://vegdri.cr.usgs.gov/viewer/) offers interactive viewing and web services. Pregenerated national, regional, and state maps are also offered at the National Drought Mitigation Center VegDRI web site (https://vegdri.unl.edu/).