Analyzing, Monitoring, and Predicting Drought Using Remote Sensing
Drought is a serious natural disaster that occurs when there is a lack of precipitation over a long period of time. The world is currently experiencing an increase in droughts, which are causing concern because they can lead to other natural disasters. The UN estimates that more than 40% of Earth’s land mass has been affected by these conditions—a number which will only rise if efforts aren’t put forth immediately towards preventing their development into something worst-case scenario. The need to identify action points for preventing drought worse-case scenarios are the basis for the development of technologies for monitoring, analyzing, and predicting drought. Remote sensing for that matter is one of these technologies that make it possible to not only analyze previous drought conditions but also to monitor current drought parameters and predict droughts in the future. Further developments by companies like Orbify have also made it easier for people who do not have a geospatial background to take advantage of earth observation techniques in monitoring drought.
Types of Drought
The underlying cause of drought is a scarcity of water, but it is further categorized into different types depending on how the scarcity occurs. The U.S. Geological Survey has categorized drought into four kinds as described below.
This type of drought occurs when the soil moisture in the root zone is either at or below the permanent wilting percentage. This type of drought continues until precipitation exceeds the daily evapotranspiration rate for that particular vegetation type.
It is defined as the degree of dryness of a particular region that’s felt during the duration of the dry period. The degree of dryness is measured against the normal or average precipitation which would have been expected for that area based on historical data during that same time frame.
It occurs when there is a rainfall deficiency that causes the water supplies resources such as reservoirs, groundwater tables, streams, and lakes to decline.
The combined effects of the aforementioned droughts result to a socioeconomic drought. This is felt in the imbalance that occurs in the supply-demand chain of economic goods such as grains, meat, fruits, and vegetables.
How to Monitor Drought
Recently, it has become possible to monitor, assess, and predict the drought conditions in a particular region using past data as well as satellite information. This allows us an opportunity to not only analyze what’s happening now but also to predict how long these droughts may last based on historical trends set against current in-situ data of weather patterns across a region. Once we’ve collected all necessary datasets we could decide on the important drought indices we would need when processing the data. Running correlations between the drought indices will help provide insight into whether there are any worrying signs ahead regarding the potential lack of rainfall over a certain period in a particular region.
The pillar of drought monitoring and prediction is data collection. We need to have relevant data before any analyses are possible. To use remote sensing for this purpose, we need various sources that provide reliable information such as rainfall data from satellites, along with onsite measurements in order to build an accurate picture of what’s happening locally before predictions can be made for different regions throughout the world. A brief description of each step of data collection is outlined below:
To understand the recent history of droughts in a region, we need to collect rainfall datasets. In the US, we could get this information from the National Weather Service (NWS). It is key to note that we need daily rainfall datasets of over twenty-thirty years for the analyses to produce a more accurate picture of drought in a region.
The next step still requires collecting rainfall data from the same region over a similar time period, but in this case, it is from satellite imagery. One option is to leverage NOAA’s AVHRR NDVI composite and the Shuttle Radar Topographic Mission (SRTM) open-source imagery for the digital elevation modeling.
The final step is to conduct a ground truth validation. For this, we need to acquire the in-situ data from the meteorological department of the concerned region. This report will show the percentage area of the region that has been affected by drought over a specific timeline. By comparing the in-situ data with the processed data, we’ll be able to tell how accurate our results are.
Drought Monitoring Indices
Many parameters ought to be taken into consideration while analyzing and monitoring drought. Drought indices play a critical role in assimilating thousands of data on water supply indicators such as rainfall, streamflow, snowpack, and others.
The drought index value is naturally a single number and so it enhances the decision-making process as opposed to using raw data. We can derive drought indices from both remote sensing and hydro-meteorological data.
Drought Indices Derived from Hydro-Meteorological Data
Hydro-meteorological data is used to recreate weather patterns that influence the world around us. The drought indices which are derived from hydro-meteorological data are continuous functions of rainfall, temperature, river discharge, and other measurable variables. The resulting number will tell us whether there will be reduced precipitation that can cause a scarcity of water. Some of the common hydro-meteorological based indices include the following:
- Palmer Drought Severity Index (PDSI)
- Balme-Mooley Drought Index (BDMI)
- Crop Moisture Index (CMI)
- Agro-Hydro Potential (AHP)
- Standardized Precipitation Index (SPI)
- Surface Water Supply Index (SWSI)
- Reclamation Drought Index (RDI)
Standardized Precipitation Index (SPI)
Among the indices listed above, SPI is the most commonly used since its analyses only require rainfall data as a base for their calculations. SPI minimizes the tedious calculation effort needed for other indices like PDSI that need more information about temperatures and humidity levels in order for their calculations.
Steps to determine the drought magnitude while using the SPI index
- Fit the long-term precipitation record at the desired station to a probability distribution such as gamma distribution.
- Transform the distribution to a normal distribution: This will push the mean SPI to zero.
NOTE: SPI values can be computed with various time scales. For instance, from one month to three months, or twelve months, and so on, until the entire time frame is covered.
The SPI index value ranges between +2.0 and -2.0. Positive SPI values denote that the precipitation is greater than the mean for that area while negative values show the precipitation is lesser than the mean. Summing up the positive SPI values for all the months in a drought event will give the drought magnitude.
Drought Indices Derived From Satellite Data
Remote sensing from space captures useful satellite data that can be used to develop drought indices for monitoring, predicting, and mitigating drought and its associated effects. Remote sensing-based drought indices are derived from data captured about the vegetation cover of an area. This is made possible by the sensors installed in the earth observation satellites which can quantify the capacity of the vegetation cover catering to photosynthesis.
When there is a drought, due to the lack of rainfall, the vegetation’s ability to carry out photosynthesis is negatively affected as the green chlorophyll pigmentation reduces. This phenomenon can be captured in the spectral response of that vegetation and visualized in a spectral reflectance curve. Observing the reflectance curve will help us know the condition of the vegetation. For instance, when the reflectance increases in the red region of the spectrum but decreases in the near infrared, it is a sign that the vegetation is stressed, unhealthy, or aging.
However, it is not only drought that could cause vegetation to be stressed. There are a number of other causes like pests, poor soils, excessive water, etc. To determine whether the vegetation is possibly stressed due to drought, we use the Normalized Differential Vegetation Index (NDVI). The NDVI reflects the percent green cover, vegetation vigor, biomass, and the leaf area index.
Normalized Differential Vegetation Index (NDVI)
NDVI is a measure of land cover that helps monitor the health and stability in an area. The data takes on different forms depending upon whether it’s being used for monitoring vegetation or hydrology, but both have been optimized by scientists to provide accurate results over large areas at low cost with high precision.
The formula below can be used to calculate NDVI:
NDVI = (λNIR – λRED) / (λNIR + λRED), where λRED and λNIR are the reflectance in the Red and NIR bands, respectively.
The NDVI index values range from -1.0 to +1.0. However, determining NDVI alone will not reflect drought or non-drought conditions, but we could discover the severity of drought by determining the deviation from its long-term mean.
Deviation of NDVI
NDVI deviation is a measure of the difference between the actual vegetation index and the expected value. It provides insight into the changes that occur in the land cover over time due to varying conditions such as drought.
The deviation of NDVI (DevNDVI) can be calculated using the following formula:
DevNDVI = NDVIi – NDVI mean, m, where NDVIi refers to the normalized differential vegetation index for a specific month, and NDVImean, m stands for the long-term NDVI for that same month.
A negative DevNDVI value denotes that the vegetation is facing drought conditions while a positive value denotes healthy vegetation. The computations can be taken to the next level of analysis to determine how close the NDVI of the month under study is to the minimum NDVI calculated from the long-term record. For this, the Vegetation Condition Index (VCI) is useful.
Vegetation Condition Index (VCI)
The VCI is an easy-to-use ratio that measures the health and quality of grasses, shrubs, or trees. It takes into account how many leaves are on each plant at all stages including new growths as well as those shedding their old chlorophyll molecules. A high VCI score means better-looking plants with adequate water supply.
The VCI is calculated using the following formula:
VCIj = [(NDVIj – NDVImin) / (NDVImax – NDVImin)] * 100, where, NDVImax and NDVImin are both calculated from the long-term record for the period under study and j refers to the index of the same period.
Unlike NDVI and SPI, the VCI is calculated in percentage. A VCI percentage of between 50 and 100 indicates that the vegetation is experiencing optimal or above-normal conditions. Any value below 50% indicates different severities of drought. A VCI value of 35% can be used as a threshold to identify extreme drought conditions. A percentage close to zero would invariably reflect an extremely dry month.
Correlation between SPI and VCI
Once the value from the SPI and VCI indices have been obtained, the next stage involves mining the data by applying association rules. First, the data is discretized to generate the association rules. To do this, we use the SPI value of the first-time scale and the monthly VCI values. As the SPI shows the rainfall deviation from the mean, the VCI quantifies the weather component of NDVI for a given location at a given time. This will lead to a firm conclusion of whether there was a drought in the study area during a specific timeframe.
Research submitted to the International Institute for Geo-Information Science and Earth Observation by Sharma (2006) used the methodology above and concluded that a discretized value of SPI that ranges from -3.0 to +3.0; wherein -3 to -1.5, is categorized as extremely dry, while -1.49 to -1.0 are categorized as dry. The other values between -0.99 to 0.99, 1 to 1.5, and 1.5 to 3, denote normal, wet, and extremely wet conditions, respectively.
Similarly, for the discretized value of VCI ( which ranges between 0 to 100); a value of 0 to 34 is clustered as very low while 35 to 45 is clustered as low. The other values, of between 46 to 65, 66 to 85, and 86 to 100 are clustered as normal, high, and very high, respectively.
Since we need to discover the drought pattern, the next step is to formulate the target episodes based on the discretized data. We could indicate that the target episodes for SPI are extremely dry and dry, while that for VCI is very low and low. Before generating rules, it is important to determine the minimum confidence to help us avoid rules that are below the confidence values. The rules can then be generated using the Apriori analysis. We can also make a choice between the set of rules depending on the interesting measure i.e. by comparing and selecting the better rules from the ones generated based on certain benchmarks.
This research by Sharma (2006) shows the correlation between SPI and VCI based on two interesting association rules.
- Extremely dry SPI leads to very low VCI [88% confidence].
- Very dry SPI leads to low VCI [71% confidence].
These rules and their confidence percentages have catered toward the correct identification of drought and non-drought years.
Rainfall and vegetation data help a long way in drought monitoring using remote sensing. From the results of research that are being carried out, we could glean the fact that the confidence in this technology is much higher than anticipated. Certainly, the combination of weather data from the meteorological department, indices values garnered from remote sensing data, and in-situ validation data would cater for the analyses to prompt drought monitoring and prediction.