Estimates of reference evapotranspiration (ETo) are no better than weather data upon which they are based. Assessments of weather data integrity and quality need to be conducted before data are utilized in ETo equations. When necessary and when possible, corrections to the data should be made to account for poor sensor calibration. Some of these corrections are described in section 1 of Annex 4.
A good cautionary statement in data analysis and application is that "no data are better than bad data." This statement applies primarily to measurements of evapotranspiration that are used to develop or to calibrate reference ET equations or to determine crop coefficients. However, it also applies to weather data. When one has no data, one can look to regional weather or ET data summaries for information that might be useful to represent conditions within the local area. In the case of ET data, one might go to a publication such as this one to make reasonably accurate estimates of ETo and ETc. However, in the case of "bad" data, meaning biased, or faulty, or nonrepresentative data collected locally, one is "stuck" with weather data and associated predictions of ETo and ETc, or with local measurements of ETc that can be biased, faulty, or nonrepresentative. The result is application of evapotranspiration data or evapotranspiration calculations to irrigation water management, to water resources operations, or to irrigation and water resources systems design that can actually cause more economic and hydrologie problems than if only reasonable estimates or even "textbook" values for ETc had been used instead. Humanity can be worse off because of faulty data as compared to no data.
Some years ago, when computer modeling was in its infancy, a common cautionary advice was to "do not trust any model until it has been validated using independent data." Today, with some of the more common mathematical models becoming proven and trustworthy, the corollary of this expression is commonly advocated, where "one should not trust any data until they are validated using a model!" Certainly, some place in between these two cautionary advocations is appropriate. Often a valid model can be valuable for evaluating data to identify errors, outliers and biases. And of course, valid data are required for selecting or calibrating a particular model.
This Annex presents guidelines to be used to calculate both extreme ranges for weather data measurements and also means to assess integrity of data that fall between the extremes. A review on instrumentation for agricultural weather stations is given first.
INSTRUMENTATION FOR MEASURING WEATHER VARIABLES 1
1 Details on weather station instrumentation can be found in FAO 27 (Doorenbos, 1976), in the WMO Guide to Agrometeorological Practices (WMO, 1981, 1983), or in meteorology handbooks (Seemann et al., 1979; Rosenberg et al., 1983; Kessler et al., 1990).
Data Acquisition and Instrumentation
Solar radiation is commonly measured with pyranometers. Pyranometers measure the shortwave incoming radiation in a solid angle in the shape of a hemisphere oriented upwards. Currently, in the most common "glassed dome" pyranometers, a thermopile is used within the 'instrument as the sensor, where thermal gradients are measured across hot and cold areas (black and white). The radiation intensity is proportional to the temperature differences between the two sensing areas. Accuracy depends upon the sensitivity of the material used in the sensors, the response time and the distortion characteristics of the material constituting the dome covering the sensors. A second type of pyranometer that is less expensive and that is gaining acceptance is the silicon diode instrument where electric current is generated by a photo sensitive diode in proportion to solar intensity. Ordinarily, silicon diode pyranometers are not fully sensitive to the full spectrum of visible light, so that the calibration of the instrument is only valid for upward solar measurements.
When a pyranometer is oriented downwards it measures the reflected shortwave radiation, and is thus called an albedometer. When two pyranometers are associated, one oriented upwards and the other downwards, the net short wave radiation is measured. The instrument is then called a net pyranometer. A point of caution is that any instrument used as an albedometer or net pyranameter must have mil sensitivity to all spectra of visible light. This is important since the composition of reflected light from vegetation is highly biased toward green. Therefore, most albedometers must be of the glass domed thermopile type and not the photo diode type.
Net radiation is measured by pyradiometers (or net radiometers), which sense both short and long wave radiation. They have two bodies, one oriented upwards and the other downwards, both covering a solid angle in the shape of a hemisphere. The sensors are made from several thermocouples sensing heat generated by radiation from all wavelengths, and are protected by domes made in general of polyethylene treated in a specific manner. The black bodies can loose their sensing capabilities with time, so that these instruments require regular and frequent calibrations. Other net radiometers are comprised of ventilated differential thermopiles, but they are very seldom utilized. All radiometers refered to above transform the radiation energy into thermal energy, a portion of which is transformed into an electric voltage gradient that provides appropriate conditions for continuous recording using dataloggers.
Sunshine duration is most commonly recorded with the Campbell-Stokes heliograph. A glass globe focuses the radiation beam to a special recording paper and a trace is burned on the paper as the sun is moving. No records occur when no bright sunshine is sensed. Measurements are reliable when the recording paper is placed in the right position according to the relative position of the sun. Care is required to avoid accumulation of rain water on the paper. The heliograph has to be oriented South in the northern hemisphere and North in the southern hemisphere. In China, another type of heliograph is used. The solar beam penetrates through an orifice and traces a recording paper treated with a sensible chemical substance. Electronic records of sunshine duration are obtained through the photo-electric or the rotating optical fibre sunshine recorders.
Windspeed is measured using anemometers, always placed at an height not less than 2 m above the ground, and often at 5 m, following recommendations by WMO. Most common are the three-cup anemometers. Also common are propeller anemometers. Measurements by both types are reliable provided that maintenance ensures appropriate functioning of the mechanical parts. Older designs of anemometers utilize mechanical counters as the output device. Modern anemometers may be equipped with generators giving a voltage signal that is proportional to the windspeed. Other anemometers may be equipped with small magnetic reed switches or with opto-electronic couplers that generate electric impulses in proportion to the windspeed. The electronic devices are utilized in automatic weather stations. Accuracy of windspeed measurements depends on the upwind fetch as much as on instrumentation. A large upwind fetch that is free of buildings and trees is definitely required for representative measurements.
The most commonly utilized sensor for measuring temperature are still the mercury thermometers. Maximum and minimum thermometers use mercury and alcohol. Bimetallic thermographs are the most common mechanical temperature recorders. They are easy to read and maintain. However, mechanical thermographs do require verification and adjustment of the position of the pen recorder. These instruments are installed in shelters that are naturally ventilated.
Modern temperature sensors have been developed, namely the thermistor and the thermocouple. These provide very accurate analogue measurements and are normally utilized in automatic weather stations. Thermistors provide independent measurements of air or soil temperature, whereas thermocouples require an additional base temperature reading, normally provided by a thermistor. To maintain the accuracy and representativeness of these instruments, they are installed in special radiation shields (shelters) having natural ventilation. Occasionally the shields or shelters are artificially aspirated to reduce biases caused by heat loading from the sun.
Dew point temperature is often measured with a mirrorlike metallic surface that is artificially cooled. When dew forms on the surface, its temperature is sensed as Tdew. Other dew sensor systems use chemical or electric properties of certain materials that are altered when absorbing water vapour. Instruments for measuring dew point temperature require careful operation and maintenance and are seldom available in weather stations. The accuracy of estimation of the actual vapour pressure from Tdew is generally very high.
Relative humidity is measured using hygrometers. Most frequently used in mechanically-based field stations are the hair hygrometers, normally operated as mechanical hygrographs. Measurements loose accuracy with dust and ageing of the hairs. Modern hygrometers use a film from a dielectric polymer that changes its dielectric constant with changes in surface moisture, thus inducing a variation of the capacity of a condensator using that dielectric. These instruments are normally called dielectric polymer capacitive hygrometers. Accuracy can be higher than for hair hygrometers. These electronic devices are utilized in most modern automatic weather stations.
The dry and wet bulb temperatures are measured using psychrometers. Most common are those using two mercury thermometers, one of them having the bulb covered with a wick saturated with distilled water, and which measures a temperature lowered due to the evaporative cooling. When they are naturally ventilated inside a shelter, problems can arise if air flow is not sufficient to maintain an appropriate evaporation rate and associated cooling. The Assmann psychrometer has a forced ventilation of the wet bulb and dry bulb thermometers.
The dry and wet bulb temperature can be measured by thermocouples or by thermistors, the so called thermocouple psychrometers and thermosound psychrometers. These psychrometers are used in automatic weather stations and, when properly maintained and operated, provide very accurate measurements.
ASSESSING INTEGRITY OF WEATHER DATA 2
2 These guidelines are based on an article by Allen (1996).
Solar Radiation using Clear Sky Comparisons
Pyranometer operation and calibration accuracy can be evaluated for a particular weather location by plotting hourly or daily average readings of solar radiation (Rs) against computed short wave radiation that is expected to occur under clear sky conditions (Rso). Rso can be computed for any day or hour as
Rso = KT Ra (5-1)
where Ra is extraterrestrial radiation 3 and KT is a "clearness" or transmission index.
3 For Ra daily computations see Chapter 3, Equations (21) - (24) and for hourly computations see Equations (28) - (33). For KT see Rso equations (3-13) - (3-20) of Annex 3.
Rso computed with equation (5-1) should plot as an upper envelope of measured Rs and is useful to check the calibration of pyranometers. Equations (3-13), (3-14), or (3-17) to (3-20) of Annex 3 should be used for predicting KT for low air turbidity. Equations (3-14) or (3-17) to (3-20) of Annex 3 are appropriate for areas with high turbidity caused by pollution or airborne dust or for regions where the sun angle is significantly less than 50°.
The example in Figure 5.1 shows one application concerning 24-hour calculations for Logan, Utah, where the highest observed values for Rs correspond to the envelope of calculated Rso, thus showing appropriate calibration of the pyranometer being utilized. In Figure 5.2, the half-hour observations of Rs for Logan are compared with the computed Rso envelopes. This figure shows good agreement between observed and computed values. However, as shown for day 7, Rs may sometimes exceed the predicted Rso when there is reflection of radiation from nearby clouds during periods when no clouds shade the pyranometer.
FIGURE 5.1. 24-hour average Rs and estimated Rso envelopes at Logan, UT during 1992 showing an appropriate calibration of the pyranometer utilized
When the Rs observations on obviously clear days fall significantly above or below the computed Rso curves, then corrective action may be warranted. The correction may be in the form of applying a correction multiplier to the observed data, so that (Rs)cor = a Rs where a is a derived correction factor. Or, an additive correction may be warranted, where (Rs)cor = Rs + c. Or, correction may be made by a combination of multiplicative and additive factors. Obviously, the corrections based on the computed Rso curves presume that the curve is accurate. The accuracy of the Rso envelope may need to be confirmed in a region by using accurate radiation measurements obtained from a calibration-grade pyranometer that has a calibration coefficient that is traceable to the international standard. The calibration pyranometer should be used only for short term periods of 10 - 14 days, and then should be stored in darkness to extend its life and to preserve the calibrated accuracy. Care should be exercised in selection of the turbidity coefficient in Equation (3-14) and (3-18) of Annex 3. Unfortunately, little information is available on this topic.
Equations for estimating hourly and 24-hour average rates of net radiation (Rn) using Rs measurements are generally accurate under most conditions. Therefore, measured Rn data should always be plotted against Rn that has been estimated using equations 4 that are based on measured Rs, air temperature and vapour pressure. The value for albedo (a) used in the Rn estimating equation should represent conditions of the surface beneath the radiometer.
4 See equations (38) through (40) in Chapter 3.
FIGURE 5.2. 30-minute average Rs and estimated Rso envelopes at Logan, UT during July 7 and July 25, 1992
If measured values for Rn chronically deviate from estimated values by more than 3-5%, then the calibration or operation of the Rn device (radiometer) should be scrutinized. This type of comparison can readily identify days or periods during which the radiometer device has malfunctioned due to effects of dust, bird droppings, moisture condensation inside the plastic domes, a lack of levelness of the intrument, or a lack of green vegetation beneath the sensor. Of course, the Rs measurement used in the Rn equations should also be scrutinized as discussed in the previous section.
The user of net radiometer data must be aware that net radiometers manufactured by different companies may not yield the same measurements of radiation even when placed over the same surface. These differences are due to differences in sensitivities of various radiometers to long wave and short wave radiation and variations among methods for calibrating sensors during manufacturing.
The type, density and height of vegetation beneath the net radiometer and relative soil moisture content should be monitored and reported with the data. Care should be exercised when positioning the radiometer to avoid shading the sensed vegetation with other instruments or structures and to insure that the radiometer is not shaded by other instruments or structures at any time of the day or year.
FIGURE 5.3. Measured and estimated Rn during 20 minute periods over mature cattail vegetation near Logan, UT during August, 1993 (from Allen et al., 1994)
Figure 5.3 shows measured and estimated Rn for cattail vegetation near Logan, UT during 1993. The measurement and calculation time step was 20 minutes. The agreement between measurements and equation estimates was fairly good. Perfect agreement between the Rn measurements and Rn equations should not be expected, due to limitations of assumptions used in the equations (e.g., the value for albedo, means for estimating the net long wave radiation component, etc.).
Soil Heat Flux
A relationship proposed by Choudhury (1989) for predicting soil heat flux density (G) under daylight conditions5 is:
G = 0.4 exp (-0.5 LAI) Rn (5.2)
where LAI is the leaf area index, exp() is the natural number raised to the exponent, and G has the same units as Rn.
5 This equation predicts G = 0.1 Rn for LAI = 2.8, which is typical for clipped grass (equation (45) in Chapter 3). Soil heat flux under forage grass during nighttime hours was found to be about 0.5 Rn. Pruitt (1995, personal communication) observed G = 0.3 Rn during nighttime hours under clipped grass at Davis, CA.
Equation (5-2) can be used to check the functioning and relative accuracy of soil heat flux plates after correcting measurements for temperature change of soil above the plates. The relationship of Equation (5-2) does not hold for 24-hour data, as a positive 24-hour soil heat flux estimate would always result. The user must be aware that Eq. (5-2) is only approximate and does not consider effects of plant spacing, sun angle, soil colour, soil moisture, or soil texture, nor the sensible heat balance at the surface on the ratio of G to Rn. Generally, more than one soil heat flux plate is used due to spatial variation in soil, soil water content, and vegetation.
Accuracy of wind measurements is difficult to assess unless duplicate instruments are used. One should always scan wind records for the presence of consistently low wind recordings. For electronic instruments, these recordings may represent a numerical "offset" in the anemometer calibration equation. The presence of these constant and consistent offsets in the data set indicates either the presence of exceptionally calm conditions (wind speeds less than about 0.5 m s-1 during the entire sampling period (which is rare)) or a malfunctioning of the wind speed sensor due to electrical shorting or perhaps due to fatigue of bearings. These problems may not be noticed by the station operator.
When possible, a second anemometer 6 of the same design, but with fresh bearings, should be placed at the weather location for a three or four day period at least once each year, and recordings compared with the permanent instrument. Variations between recordings can signal a need to replace bearings, switches, or other parts.
6 If a second data logger is used to record the temporary anemometer, one should be careful to synchronize data logger clocks. Also, one should be careful that anemometers do not interfere with one another's wind stream.
Relative Humidity and Vapour Pressure
Vapour pressure of air is difficult to measure accurately. Some older electronic humidity sensors were commonly plagued by hysteresis, nonlinearity and calibration errors. Some of these errors are inherent in the sensor design and still plague some modern sensors. Other errors result from dust, moisture, insects, pollution, and age.
Replication of RH Instruments
It is strongly recommended that duplicate RH and air temperature sensors be permanently employed in electronic weather stations, at least for some period each year. When duplicate RH and air temperature sensors yield similar measurments, then it is likely that both sensors are functioning properly, provided proper calibration equations have been used. However, even though duplicate sensors are in agreement does not mean that the readings are free from calibration errors and biases due to nonlinearity, etc..
Trends in Computed Dew Point Temperature with Time
When air humidity is measured using RH sensors, the actual vapour pressure of the air (ea) is calculated as:
where e° (T) is the saturation vapour pressure at air temperature T and RH is in %. RH and T must be taken for the same time period, preferably for £ 1 hour.
Hourly (or shorter) measurements of RH, dew point temperature (Tdew) or vapour pressure (ea) can be preliminarily assessed by plotting hourly measurements of computed Tdew or ea with time. Relative humidity will vary significantly with time of day, and inversely with air temperature as shown in Figure 12 of Chapter 3. However, both Tdew and ea, either measured directly, or computed using RH and T measurements, should remain somewhat constant throughout a 24-hour period when the air mass is stable and advection of dry air from outside the area does not occur. During these stable periods, one should expect some rise in Tdew and ea during daytime periods, when ET fluxes humidify the equilibrium boundary layer. However, this increase is usually less than about 10 to 20%. Variation in Tdew increases significantly when a weather front passes overhead. Since ea is calculated as the product of RH and saturation vapour pressure at air temperature, any error in the RH calibration tends to cause false variation in Tdew and ea with changing air temperature.
Figure 5.4 shows Tdew computed from measurements of RH and air temperature at a weather station in the center of a wetland near Logan, UT (20-minute data). Tdew generally varied from hour to hour due to air mass instability and increased during most days of this period as evaporation from the local wetland vegetation added humidity to the air. The data sequence shows some periods of relatively constant measurement (calculation) of Tdew throughout a 24-hour period (for example day of year 199), even though air temperature varied substantially. This is a good indication that the RH sensor was probably functioning correctly and that the instrument calibrations were probably valid.
Figure 5.4 also shows, for the same weather station, a comparison between RH measured using two different and independent relative humidity sensors. The two sensors, one a "chilled-mirror" device that measures Tdew directly, and the other, a device that measures RH directly, agreed very well with each other during the 8 days shown. The value of having "redundancy" in instrumentation is demonstrated in this example, where the two different devices measuring the same parameter (in this case RH) leave no question concerning the validity and accuracy of the RH measurements, due to the close agreement. The use of only a single instrument would leave some question as to accuracy.
One can notice in Figure 5.4 that the RH approached 100% on day 200, which is expected for a well-watered setting. The difference between minimum daily temperature and Tdew was generally 1 to 2 °C for many of the days. This is expected in dry, advective environments, as discussed in Chapter 3 and Annex 6.
Observations During Periods of Dew and Rainfall
In many climates, especially those where nightime dew occurs, Td during early morning hours before sunrise should coincide closely with recorded Tmin and RH should approach 100%. For automatic recording weather stations where recording rain gauges are used, one should expect RH recordings during periods of rain or light drizzle to exceed 95 %. Relative humidity recordings that exceed 100% by more than 3-5% during early morning hours or during precipitation events indicate a need for recalibration and numerical adjustment of collected data.
FIGURE 5.4. Tdew and RH from measurements near Logan, Utah, the United States during 1995 (20-minute data)
Maximum Daily Relative Humidity
When humidity data are measured in a reference setting, early morning RH will often approach 100%, even in semiarid areas if measurements are taken inside an irrigated region. Values of maximum relative humidity (RHmax) that consistently fall below 80% to 90% when in an irrigated or well-watered setting may indicate problems in RH sensor calibration or functioning or may indicate aridity of the measurement site and deviation from reference conditions.
Figure 5.5 shows daily measurements of RHmax from an electronic agricultural weather station located near North Baltimore, Ohio over a five year period. One would expect RHmax to approach 100% in this subhumid setting. However, one can see clear evidence in Figure 5.5 that the RH sensor was undermeasuring RHmax during several years, with decreasing trends in RHmax visible during these years. This indicates that the RH sensor was functioning electronically, except during the first half of 1988. However the calibration of the sensor element had seriously decayed and was not valid for 1988, 1990 and 1992. Sensor elements were typically replaced in September of each year. RH data for 1990 and 1992 could potentially be corrected by multiplying the RH measurements by a correction factor or by adding an offset.
FIGURE 5.5. Daily values of measured RHmax at North Baltimore, Ohio (1988-92) showing inappropriate calibration of the sensor for 1988, 1990 and 1992
The type of plotting and screening demonstrated in Figures 5.4 and 5.5 shows the simple types of integrity assessments that can be utilized in near-real-time or with historical data. These types of assessments can be applied to all weather data used in evapotranspiration estimation and should be adopted by operators of agricultural weather networks.