Requirements for environmental data quality in farming – case from IPM 9.12.2018 Requirements for environmental data quality in farming – case from IPM Hanna Huitu MTT Crop Production Technology Team MMEA certainty seminar 6.11.2013 MIKES, Espoo 9.12.2018
What environmental data is being collected? 9.12.2018 What environmental data is being collected? Background data: field plot registry, soil information, cultivation history.. Data that varies in time: Temperature, precipitation, soil moisture, state of vegetation Ways to collect data: Weather stations & other weather data Sensing from machinery In situ monitoring probes (soil moisture) Aerial photos, UAV Data pool in use. Bacground data is requested by legislation, at parcel level. Forms the framework that is in use. Open data Sensing from machinery – Nitrogen sensing, GreenSeeker © Maa- ja elintarviketalouden tutkimuskeskus 9.12.2018
Drivers for Environmental monitoring in cultivation EU requirements Demands for food production chain, Food traceability Ensuring social acceptance Information management and planning at the farm Process control in agricultural operations Monitoring of soil Phosphorus and Nitrogen levels is a compulsory part of EU Agro-Environmental subsidies. Agri-environmental payments are granted for voluntary commitments going beyond good farming practice. Chemical pest control is regulated fairly strictly Observations—weather, soil moisture, vegetation stress Decision support – weather-based harvest time estimate for grass Use in automation – pesticide spraying takes into account wind when adjusting the nozzles of the spraying system Observations decision support use in automation Observation data in use Modeling results in use
Environmental monitoring in practise Environmental measurements are a part of information management in the farm Data use and transfer is evolving and increasing Farm machinery is one piece of equipment in data aquisition Often measurements need to be geo-located Measurements provide support for decision making Farm machinery one part in data aquisition: nitrogen sensors (weather services, frost alarms, risk estimates for plant diseases, harvest time forecasts)
Role of weather data in agricultural cultivation 9.12.2018 Role of weather data in agricultural cultivation Farming relies on weather Weather based decisions involve activities that should happen in the very near future (<week) Both observations and forecasts From science to operational applications: Weather data is utilized in model development Models are then used in applications that are fed with weather data Primary driving variables in biofysical processes and determining management responses. Examples: irrigation, freeze protection, pesticide spraying, and harvesting © Maa- ja elintarviketalouden tutkimuskeskus 9.12.2018
Availlability of weather data 9.12.2018 Availlability of weather data Not dependent on only one source - examples: Automated “amateur” weather stations on field FMI routine observation stations Grid data, based on FMI observations and modeling LAPS data Weather forecasts Here an example of the use of interchangeable data in ”system of services” framework, where real-time weather data sources (observation data or forecasts) are connected to modeling engine as services, and user can select the weather data source. LAPS = Local Analysis and Prediction System SOURCES ARE DIFFERENT! Data quality comes into play. © Maa- ja elintarviketalouden tutkimuskeskus 9.12.2018
Location specific weather data for fields SoilWeather network: 49 monitoring stations, owned and operated by MTT and SYKE. Temperature, relative humidity, precipitation, wind also water quality, discharge, soil moisture Set-up in representative locations Pre-scheduled maintenance and maintenance according to the need Daily QC with simple algorithms, built and operated by SYKE Information on the problems is delivered to the maintenance daily Each value in the data is supplied with a quality flag Variation of data quality can be caused for example by problems in data transfer, sensor fouling, other disturbances, natural conditions
9.12.2018 ms36 Each plot has a location for a weather station in the middle. Quality that can be reached by farmers on an amateur weather stations. Questions about how representative is the rainfall to the field, given the wind. Comparison: Plotted how much does hourly temperature deviate from the temperature measured at closest by FMI station. © Maa- ja elintarviketalouden tutkimuskeskus 9.12.2018
Location specific monitoring on fields 9.12.2018 Location specific monitoring on fields Soilweather weather stations show ~0.1 C bias for temperature and 0.2% for relative humidity when situated next to FMI stations (cc: 0.99 and 0.94 respectively) With longer distance between weather stations, bias of up to 2 degrees in temperature was not unique in the area. Deviations during one growth season match the deviations on another growth season Conditions for weather-dependent processes are field specific Favours use of in situ monitoring on the field plot, or mathematical modeling. Elevation explained some. Crop growth starts when temperature in top soil reaches © Maa- ja elintarviketalouden tutkimuskeskus 9.12.2018
Does source of data matter? 9.12.2018 Does source of data matter? Compared three weather datasets in relation to model results and their recommended actions (nearest FMI station, FMI modeled data grid, soilweather weather stations) Disease risk prognosis models: Need for these models is in increase, and new services are currently in development Disease risk prognosis is now piloted as a new feature in crop planning system software Soilweather weather stations are also used in (some of the) plant disease trials and development of risk model. Now focus on how selected data sources perform when model is utilized in multiple field locations To get an idea how different data works for an application © Maa- ja elintarviketalouden tutkimuskeskus 9.12.2018
Implementation of the weather data in disease risk modeling 9.12.2018 Implementation of the weather data in disease risk modeling Cereal leaf spot disease development Real-time risk level and suggested timing for pesticide sprayings Combines local weather information and cultivation information (crop planning system) Accurate estimates save both money and environment. Use of chemical protection is now allowed only when needed. Existence of the need could be proved by models. Here we use as an example crop disease models. We have dealt with DTR and tan spot
Does data accuracy affect decisions? 9.12.2018 Does data accuracy affect decisions? Uncertainty in humidity changes the steepness of the risk accumulation curve, but there is no significant effect on the timing of the accumulation. Uncertainty in temperature changes strongly the timing of the beginning of accumulation, and slightly the steepness of the accumulation curve. Both have an effect on the suggested pesticide spraying, which is carried out as when specific risk level is reached a X axis is time. Y-axis is risk level. Verkkolaikku – ruskolaikku -- pistelaikku b Cumulated risk value based on a) in situ measurement and b) data from FMI stations 9.12.2018
Did data quality affect decisions? Model was run with alternative data to provide risk cumulation curves On some of the fields, estimations were validated by field sampling of vegetation Calculated risk levels fit nicely the observations of disease occurrence Both local weather measurements and FMI GRID data produced good estimates on the disease pressure, and also use of nearest FMI station was still leading to recommendations that were of good use by expert judgement © Maa- ja elintarviketalouden tutkimuskeskus 9.12.2018
How did we see data quality? 9.12.2018 How did we see data quality? Measurements should describe the thing we want to measure. (actually this model would need leaf wetness information, but we used relative humidity, precipitation and temperature) Data is applicable, regular and well-described for its intended use: information is useful Value of data actualizes through its use. Uncertainties carried by models affected end result more than uncertainties in source data (even epidemiology of the disease may change season to season, models give only approximations) Acquisition of improved data is often not free of cost - efficient solutions tolerate certain level of input data inaccuracy Ending up to fairly philosophical question on what is it that we want to measure? © Maa- ja elintarviketalouden tutkimuskeskus 9.12.2018
Data quality in cultivation? 9.12.2018 Data quality in cultivation? Monitoring data and related real-time model results are one part in knowing what is going on - a picture that is constantly being gathered from multiple channels Value of data can be seen in better decisions that will result (cost - benefit). If result is dependent on the data source, comparison based on result can be fruitful Operational real-time applications require robustness from weather data sources. Model development can benefit from accurate, maintenance-intensive weather data sources. Science – logic: create a model based on a data sample, and extrapolate that to operational use. Awareness of the situation. © Maa- ja elintarviketalouden tutkimuskeskus 9.12.2018