2.1. Karjaanjoki river basin
SoilWeather is an operational river basin scale in-situ
wireless sensor network that provides spatially accurate, near real-time information on weather conditions, soil moisture and water quality with a high temporal resolution all-year round. The network was established in Southern Finland during the years 2007 and 2008 and it covers the entire 2,000 km2
Karjaanjoki river basin which is located in south west Finland (Figure 1
). The catchment is mainly covered by forest (63%) and agricultural areas (17.7%). In the north part of the area the River Vanjoki and River Vihtijoki bring waters to Lake Hiidenvesi (area 29 km2
, mean depth 6.7 m) from which waters flow via River Väänteenjoki to Lake Lohjanjärvi (area 92 km2
, mean depth 12.7 m). Finally, the Mustionjoki river transports water from the river basin to the Gulf of Finland. In the northern parts of the river basin geology is dominated by quartz and feldspar. In the south the bedrock is granite. The soil is mainly clay, silt and glacial till [23
The weather stations are evenly distributed around the catchment (Figure 2
). They serve the purposes of catchment wide run off modeling. The turbidity and soil moisture sensors are scattered around the catchment as well, still majority of them are placed on the areas of different applications, which are explained later. Specific nutrient measurement stations are placed totally on the local application areas.
There are three intensively measured areas within the river basin: Hovi farm, Vihtijoki sub-catchment and Lake Hiidenvesi (Figure 1
). The sensors are mainly located on land owned by private farmers, who are also the main users of the data. Eleven of the weather stations are placed in or close to potato crops for potato late blight warning. In addition data from one weather station close to a potato late blight control experiment at Jokioinen outside the SoilWeather network was used to evaluate the validity of potato late blight forecasts. The water measurements are obtained mainly in the rivers, but also in relatively small ditches and in constructed wetland within the Hovi intensive measurement area.
In the Hovi farm (25 ha) we measured soil moisture, weather and water quality at a field parcel level. The Hovi farm in Vakola is owned by the governmental MTT Agrifood Research Finland research institute. The soils are mainly clay, silt and glacial till and altitudinal variation is low (up to 130 m). Crops include barley, grass, turnip rape and wheat. Constructed wetland was built at Hovi farm in 1998 for water treatment, biodiversity and landscape purposes. The catchment of the wetland (12 ha) is under cultivation. It is a relatively large constructed wetland, ca. 5 % of the whole catchment [24
]. One turbidity sensor is installed in the middle of the wetland and two spectrometers measuring nutrient concentration are located in the inflow ditch and close to the mouth of the outflow ditch of the constructed wetland to monitor its effectiveness in nutrient retention. Additionally, there are five weather stations in the area of the Hovi farm. (Figure 3
).The spatially dense instrumentation of Hovi enables monitoring and testing of water protection methods and management practises and studying nutrient leaching from agricultural land in varying weather conditions at the field parcel level.
The Vihtijoki sub-catchment, located in the north-west of the Karjaanjoki river basin, is instrumented with 25 weather stations and six water turbidity sensors. The turbidity sensors are located in the upper, middle and lower parts of River Vihtijoki to obtain validation data for the modelling efforts on transport of phosphorus (P) and total suspended solids (TSS) at a catchment level. Here, the SWAT (Soil and Water Assessment Tool) model will be used. SWAT is a catchment-scale model that operates on a daily time step [25
] and simulates water and nutrient cycles. SWAT model needs time series of many weather parameters as a part of input data. We intend to test the sensitivity of the model to the frequency of weather stations, i.e. if, and how much, will the results improve when the number of the weather stations will be increased in the model-setup, which will be established in the Vihtijoki sub-catchment (Figure 1
Lake Hiidenvesi is one of the largest lakes in Southern Finland. It has recreational importance and it serves as a backup drinking water reserve for the inhabitants of the Finnish capital area. Restoration of the lake was started already in 1995 due to low water quality but improvements in water quality have not been gained so far . The SoilWeather WSN has been used to monitor the water quality of the inflow and outflow of the lake using two nutrient measurement stations and one turbidity sensor.
2.2. Sensors, sensor network and infrastructure
The Soil Weather WSN hosts 70 sensor nodes altogether; 55 compact weather stations, four nutrient measurement stations, and 11 turbidity measurement stations. Six of the turbidity stations have water level pressure sensors as well. The typical setup of a weather station includes a weather station core and sensors for air temperature, air humidity, precipitation, wind speed and wind direction. Connected to the weather station cores there are also sensors for soil moisture and for water turbidity so that the network observes in its entirety soil moisture in 30 sites, turbidity in 18 sites and water level in eight sites.
Nutrient measurement stations measure water turbidity and nitrate concentration with spectrometers employing ultraviolet and visible (UV-Vis) wavelengths. The setup includes also sensors for water level and temperature. All the sensor nodes have been geo-located in the field using a hand-held GPS device Trimble GeoXT. The sensor nodes, sensors and parameters measured are shown in Table 1
SoilWeather WSN uses off-the-shelf sensors, nodes and server services provided by various sensor vendors. Each sensor node has a central processing unit with a GSM modem and SIM-card installed either into a weather station core or into a nutrient measurement station. The weather station cores can be controlled remotely by SMS messages or locally by connecting sensor nodes to the computer. The cores can also be programmed to produce automatic SMS alerts e.g. on drought, frost or moisture conditions predisposing to plant diseases.
The network uses time-based data collection. The frequency for nutrient measurements is once every hour, all the other sensors measure once every 15 min. Each sensor node collects and transmits the data independently to the database server, either as a SMS message (a-Weather station cores) or as a data call (nutrient measurement stations). The weather station cores are wireless and automatic; GSM and GPRS techniques are used in the data transfer and storing. GSM modems receive SMS messages, GPRS messages are transferred through HTTP interface. These messages are written to a message database and decoded with a parser program to measurements and timestamps. This information is then written to the final database.
The near real-time data is available as graphs and downloadable tables in two different internet-based data services provided by the sensor vendors. One of the services also supports XML-based data transfer. Diagram of the data flow in SoilWeather WSN is presented in Figure 4
The SoilWeather WSN functions all-year round. Due to freezing of the sensors, measurements are less accurate in cold winter times, as there is no heating in the rain gauges or wind sensors. Also, the sensors located in rivers may be temporally removed during winter, as moving ice might break the sensor probes.
The weather stations are compact devices including all the sensors installed and they are easy to deploy. The weather stations are programmed to connect to the server automatically. For water turbidity and nutrient measurements, the easiness of deployment is very much dependent on environmental conditions, such as the ground material of the river bank and river bed, the river run-off, and existence of constructions. The sensor nodes are transferring data independently, and network is flexible to some extent; it does not demand reprogramming or updating of the existing nodes when new node or sensor is added. The sensor nodes use a battery package of two 6 V batteries.
At the moment, the data for the whole network is available only for participants of the project. The weather measurements are, however, freely available for the previous month through the open interface at the web site http://maasaa.a-log.net/
(in Finnish) and through the web site of Helsinki Testbed ( http://testbed.fmi.fi/
) after registration to researcher’s interface.
2.3. Data quality control and network maintenance
We see data quality as a broad concept including aspects of deployment, maintenance, cleaning, calibration and automatic data quality control algorithms. Careful deployment of sensor probes is the basis for ensuring good data quality. The location of the probe should be representative, considering the parameter measured. Weather stations are located in open and relatively flat areas and water turbidity sensors in the main run-off in location with no nearby discharging ditches or tributaries. The probes are mainly deployed by the same experienced field assistants from nearby MTT Vakola farm and by following sensor specific procedure. However, the final location of the sensor probes was always decided by the application, and negotiations with the land owners. The probes are also located so that they do not hamper cultivation practices or the recreational use of the river.
All the water and soil sensors are calibrated against water or soil samples, respectively. For weather stations no calibration in the field is done. Calibration samples for water measurements are taken once a month to ensure the quality of the sensor measurements and the correct functioning of the sensors. River discharges are available close to the location of the water measurements. Soil moisture calibration samples were taken soon after the deployment.
Reliable functioning of the sensors requires maintenance often enough. We maintain sensors on a regular basis, twice a year, but also occasionally when additional maintenance is needed. The maintenance procedure is sensor type specific. For weather stations the batteries are changed once a year, the fixation of instruments is checked and fixed if needed, and the equipment is cleaned. The water turbidity sensors and nutrient measurement stations need extra care because the optical lenses get contaminated in the water. The spectrometers are cleaned automatically with air-pressure and in addition manually once a month. Some of the water turbidity sensors are equipped with automatic wipers. The wipers were not available during the first deployments so the sensors were manually cleaned in regular basis: in winter time every month and in summer time when needed, approximately once a week.
Automatic data quality control system, that warns when suspicious data is received, was developed during the project to notify on maintenance needs. Different kinds of data quality problems that can occur in the SoilWeather WSN data are shown in Figure 5
. In the first chart (a) there are two suspicious spikes in the temperature data. Rather common situation of missing data is shown in the second chart (b) and in the third chart (c) the wind speed is for some reason measuring the same value (0 m/s) all the time. In the beginning of the project there were only a few stations providing data to be checked and the quality control was carried out manually. As the amount of stations, and therefore the amount of the data, grew, it was essential to develop an automatic quality control and warning system. At the moment the system checks the data from all the a-Lab sensor nodes. For the four nutrient measurement stations Luode Consulting handles the quality control manually using their strong expertise and experience in this field.
The automatic quality control runs under the UNIX system. The computer of the data quality controller logs in to the a-Lab server via SSH tunnel and retrieves data using Matlab Database Toolbox. After the tests are run in Matlab, the quality controlled data is returned to a new database in a-Lab server. At the moment there are four different tests running in near real-time:
The missing data test checks if the data has been sent correctly. If no observations have come from the sensor within the period after the last check, the system saves an error report. Meanwhile the missing data test checks long periods of missing data, the second test searches for occasional missing values. The third test is for checking if the measurements vary over time. Presumably there is something wrong with the station or the sensor if the sensor measures the same value consistently (for 24 hours in this case). Finally, the range test tests if the measurement lies between predetermined range values. For meteorological parameters limit values were configured based on seasonal climate extremes and limit values vary according to the month and the climatic zone. The climatic zone of the Karjaanjoki river basin is hemiboreal and the range values in this case for air temperature are shown in Table 2
. Limit values for meteorological parameters are provided by the Finnish Meteorological Institute (FMI). Soil humidity, turbidity and water level ranges are defined for every sensor separately, depending on the characteristics of soil, riverbed and river hydrology. For every observation the system gives an information label (flag) that indicates the quality level of the observation according to the range test. The flag value indicates whether the observation is correct (between the range values), suspicious (differs slightly from the range value) or wrong (differs dramatically from the range value). The range test and the flagging follows the system used in FMI [28
All the error messages from the past 24 hours are collected and sent automatically by e-mail to the data controller every morning. After the notification, the controller checks the data manually and makes the decision weather to inform the maintenance team or not. All the maintenance and the cleaning activities are stored in the log file of the sensor node and the log file is available for users through the data services.
SoilWeather WSN is designed to be a multi-functional network. During the two-year pilot project, it has been utilised in the following applications:
in predicting potato late blight risk
in developing interpolation methods for weather parameters into 30 m resolution grid
in monitoring water quality and nutrient retention in rivers and in constructed wetland
in improving hydrological model at river basin scale
in leaching model in sub-catchment scale
in soil moisture model at field parcel level
in precision agriculture.
It has also been used to study the relationship between local weather conditions and nutrient leaching. The network enables monitoring weather-related phenomena, such as heavy rains and the nutrient load peaks they induce. The SoilWeather WSN is used in research and in governmental monitoring tasks, but also by private farmers, who can use local data in planning and executing management practices. Here we present and analyze two applications in detail: predicting potato late blight risk in the farms, and the monitoring of constructed wetland.
Potato late blight caused by an oomycete, Phytophthora infestans
, is one of the most devastating potato diseases worldwide. The potato crop can be completely destroyed within a few days if the weather is conducive for disease progress (Figure 6
). In modern conventional potato production late blight can be effectively controlled with a range of chemical fungicides. The potato crop must be protected from emergence to harvest for each single day when weather enables late blight infection. Fungicide applications are necessary at 3 – 10 days intervals throughout the growing season resulting in 4 – 10 consecutive sprays in Nordic production and more than 20 sprays in the most intensive potato production regions in Western Europe [29
To optimize the number of fungicide applications per season numerous weather based blight forecast models have been developed since the 1950s [31
]. In the Nordic countries a late blight forecast model (NegFry) developed by Fry et al
] has been widely used since the 1990s [13
]. Dramatic changes in the epidemiology of potato late blight pathogen have made the old NegFry model unreliable in certain occasions [33
]. Therefore a more recent potato late blight model (LB2004) introduced by Andrade-Piedra et al.
] has been modified for use in the Nordic climate [35
]. The characteristics of current Nordic potato late blight populations for the model development were studied in detail [36
] and essential epidemiological parameters needed in the model were updated [37
]. Sub-model calculation periods when the temperature is over 8 °C and relative humidity is over 90 % [35
] was used to predict blight risk in this study.
The blight risk was calculated for the 11 weather stations at the potato fields and at the weather station at late blight control experiment at Jokioinen. The potato fields were visited twice a week from the last week of June to the first week of August. The occurrence of potato late blight was recorded and the onsets of blight epidemics were reported in the Web-Blight warning service ( www.web-blight.net
). The severity of blight as a percentage of defoliated leaf area was assessed at the experiment at Jokioinen three times a week.
Constructed wetland studies were made during 1999–2002 at the previously mentioned Hovi wetland [24
]. As for the monitoring of water quality of inflow and outflow, the measurements were based on water sampling. Although the sampling earlier was flow-proportional and rather frequent, most of the days were left unmonitored. However, these days may include short-termed peaks of high runoff, which remain unknown. Typically, the gaps between the sampling days have been filled by e.g. linear interpolation, but the loading estimates tend to be more or less erroneous. Flow variations and thus also the error is particularly significant in small, agricultural, high-sloped catchments like the Hovi farm. For this defect, automatic sensors providing non-interrupted data offer a revolutionary improvement. To test this new monitoring approach in wetland research, s::can -sensors (Table 1
) were installed in October 2007 for monitoring of the water entering and exiting the Hovi wetland at 1-hour interval. The first full 1-year results (from November 2007 through October 2008) on the retention performance of the wetland were compared with the previous, water-sampling –based results [38