The key question when looking for observed weather data: How do they generate weather data for areas between ground observation stations?
This is not a trivial question. For Europe and USA, the ground station observation network is dense in populated areas to support the generation of accurate gridded weather observations. But what about other regions with a low density of weather stations such as Africa, Central and South America, and South Asia?
aWhere offers the premium, global, daily observed weather data dating back to 2006. These data have stationarity and are statistically robust to support a range of economic sectors in service of informing adaption to climate change. Examples include agriculture, water, health, logistics, infrastructure, and security – to name just a few. There are many weather companies and there are significant differences in the methods each uses to generate historical observed weather data.
This blog highlights critically important differences in how the historical weather is created.
Observed weather? There are a few key words utilized to describe historical weather – and note that here I did not say ‘observed’ weather. When the method is based on techniques such as simulated, re-analysis, zero hour forecast or of late ‘machine learning (ML)’ or ‘artificial intelligent (AI)’ algorithms then a question needs to be asked as to the input used for these methods. Typically, the input to all these methods is the traditional ground station. It only follows that any single ground station will have a greater contribution to the resultant gridded surfaces in areas where there are fewer ground stations.
For data scarce areas, this is an enormous challenge. Many variables can be accurately interpolated ‘between ground station observations’ and especially so if the elevation is obtained from any of a variety of high-resolution elevation data sources. Using elevation, one can accurately interpolate temperature and relative humidity. Precipitation, however, cannot be interpolated. Rainfall is a discontinuous variable. We all have the experience of watching thunderstorms roll by pouring rains along a path that might only cover a swath of 15 kilometers. The edge, where the storm misses, can be quite abrupt. We know this because areas with dense ground station networks confirm this observation. As a result, daily gridded rainfall presents a wholly different challenge with the accuracy of such interpolated efforts being dependent on ground station density.
Model-generated daily grids based on ground stations to create all weather variables runs into accuracy issues for regions with few operational ground stations resulting in inaccurate rainfall grids. No matter the approach to generate a rainfall grid based on too few ground station observations, these simulations cannot accurately provide rainfall between ground stations.
The map below of Northeastern Mozambique shows the location of four ground stations bounding an area of over 40,000 square kilometers. The ground observations all report above normal or normal rainfall. aWhere uses satellite derived rainfall observations for our daily rainfall and for this area bounded by the 4 ground stations we offer more than 500 observations of rainfall. aWhere’s weather station grid shows that it was actually much drier than normal ‘inside’ the area covered by four ground stations. There are no interpolation methods, ML or AI, that would have placed drier-than-normal conditions bounded by these four ground stations.
Data scarce areas, particularly sub-Saharan Africa, Asia, and Latin America, are home to a large population of Small Scale Producers (SSP) who farm under 5 acres or 2.5 hectares. To support SSPs in these times of increasingly variable weather, digital agronomics offer tremendous potential to empower farmers to adapt to climate change. For agricultural advisories and recommendations, accurate observed weather data are required to drive crop models that enable stage-specific recommendations to farmers. Accurate forecasts are also needed to help farmers plan farm activities but translating a forecast into an agronomic insight requires crop, pest and disease modeling. With accurate observed weather data, crop growth stage, soil moisture status and other key agronomic characteristics can be modeled. SSPs can receive timely and targeted crop advisories via text message to optimize farm operations, reduce risk of crop failure and ultimately make their farm more profitable. For example, at a specific crop growth stage and with good soil moisture from rains, the forecast for additional rain may trigger a recommendation for additional top-dress fertilizer. Under these environmental conditions, fertilizer has a high ROI (Return on Investment). In the absence of accurate observed weather data, farmers would not likely have full confidence in the forecast alone to apply additional fertilizer.
Increasingly, rainfall events are becoming more extreme due to a warming atmosphere. “Global warming increases both the intensity and frequency of extreme precipitation, so to characterize the full response of extreme precipitation to global warming, either their total or both of their individual contributions must be communicated.” (Myhre et al., Nature, 2019) Generating accurate rainfall grids is of paramount importance to develop adaption strategies associated with documented increase in rainfall variability and intensity. For agriculture, accurate, localized, and timely rainfall data is a critically important weather variable. Rainfall cannot be interpolated and any simulated or re-analysis product by definition will not capture the ‘tails’ of rainfall distribution associated with extreme rainfall events. From the same article just referenced: “Further, observations indicate that the total precipitation from extreme events occurring once per decade may increase on the order of 10 times more than when considering intensity increases alone.” aWhere understands the limitations of simulated weather data and has addressed this to offer accurate, localized, observed daily rainfall data for the world. This is our differentiation.
aWhere has used advanced data science to blend remotely sensed weather data and ground station data to generate 1.9 million complete virtual weather stations every day. These resources now empower a wide range of clients to assess weather risks, especially the agriculture sector. aWhere uses state-of-the-art data science to interpolate weather variables like temperature, using 3D curvilinear functions to adjust for elevation between ground stations and applies a statistically consistent methodology over the entirety of aWhere’s historical weather record. This enables aWhere to use advanced algorithms to detect faulty ground stations and eliminate faulty sensors for a given variable on a per location per day basis. We do statistical checks each day for every single variable for each one of the 1.9 million virtual weather stations in our 5-arc minute (about 9km) grid. With aWhere, there is no interpolation of rainfall – only observed rainfall which leads to greater accuracy and application in our increasingly variable world of weather.