天气的错误测量：在经济背景下使用遥感地球观测数据（英）

11015 The Mismeasure of Weather Using Remotely Sensed Earth Observation Datain Economic Contexts Anna JosephsonJeffrey D. MichlerTalip KilicSiobhan Murray Development EconomicsDevelopment Data GroupJanuary 2025 Policy Research Working Paper11015 Abstract The availability of weather data from remotely sensed Earthobservation data has reduced the cost of including weathervariables in econometric models. Weather variables arecommon instrumental variables used to predict economicoutcomes and serve as an input in modeling crop yieldsfor rainfed agriculture. The use of Earth observation datain econometric applications has only recently been metwith critical assessment of the suitability and quality ofthese data in economics. This paper quantifies the signif-icance and magnitude of the effect of measurement error in Earth observation data in the context of smallholderagricultural productivity. The paper shows that differentEarth observation sources use different measurement meth-ods. The findings are not robust to the choice of Earthobservation dataset, and the outcomes are not simply affinetransformations of one another. Thus, the paper suggeststhat researchers should exercise caution in using these dataand include robustness checks that test alternative sourcesof Earth observation data. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about developmentissues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry thenames of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely thoseof the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank andits affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. TheMismeasureofWeather:UsingRemotelySensedEarthObservationDatainEconomicContexts∗ AnnaJosephson1,JeffreyD.Michler1,TalipKilic2,andSiobhanMurray 1DepartmentofAgriculturalandResourceEconomics,UniversityofArizona2DevelopmentDataGroup,WorldBank JELClassification:C38,C81,D83,O13,Q12Keywords:RemoteSensingData,SocioeconomicData,MeasurementError,Weather,Sub-SaharanAfrica Science must be understood as a socialphenomenon... not the work of robots programmedto collect pure information. Scientific findingsshould not be elevated to the status of immutabletruths. —Stephen Jay Gould, The Mismeasure of Man 1Introduction There are 37 weather stations in the World Meteorological Organization’s (WMO) database on theentire continent of Africa. These stations provide weather data for 1.1 billion persons living on thecontinent (Tzachor et al., 2023). This figure stands in contrast to the 636 weather stations acrossthe European Union and the United States, which provide weather data for 1.2 billion personsin those two regions (Tzachor et al., 2023).Further, because of their uneven dispersion acrossthe African continent, the weather stations that exist cover only about 40 percent of the Africanpopulation. Weather stations on the continent are often so far apart that the data collected are oflimited use, a condition exacerbated by under-investment in their maintenance, which results in adeterioration of the frequency and quality of data reporting. Only one in five weather stations inAfrica met the WMO’s reporting standards as of 2019. The lack of station data across Africa means that, for much of the continent, the truth ofthe weather is unmeasured.The emergence of Earth Observation (EO) and improvements inweather modeling have enabled the development of remotely sensed weather datasets to fill thisgap. These products provide an estimate of the on-the-ground conditions of weather, albeit detectedfrom a distance.These EO products can provide myriad measures of weather phenomenon, likeprecipitation, temperature, wind speed, or humidity. Each EO product uses a different combinationof weather sensors and methods for interpolating and interpreting the data from those sensors. Andso, while each EO product endeavors to measure the objective truth of weather conditions, eachproduces its own “truth.” In theory, these “truths” should all be the same. In practice, they arenot.Figures 1 and 2 show these different “truths” across six remotely sensed EO precipitationproducts and three temperature products. One precipitation product reports rainfall of less thanfive mm while a different product reports rainfall of more than 47 mm for the same location on thesame day. Temperature also varies by EO product. One product reports a maximum temperatureof 23◦C while another reports the maximum temperature that day as 27◦C. Where does this disparity leave researchers who want to incorporate EO weather data intoeconomic contexts?If using ground-based weather station data, the gold standard